Split (1)

train · ~5.33M rows (showing the first 168k)

id string	text string	source string	added string	created string	metadata dict
PMC521687	Background ========== There is increasing evidence that sympathetic nervous system mediated cardiovascular stimulation with increased catecholamine blood levels is the principal mechanism responsible for perioperative tachycardia and hypertension, myocardial ischemia and infarction \[[@B1]-[@B3]\]. Even short-lived changes may have detrimental effects on the coronary circulation of high-risk patients, with higher rates of morbidity and lethality \[[@B4],[@B5]\]. Thus, many studies have concentrated on the stressful stimulus of endotracheal intubation, and a number of pharmacological attempts have been used to attenuate the hemodynamic response, including the use of high doses of opioids, α~2~-adrenergic receptor agonists, β-adrenergic blocking drugs or other antihypertensive drugs \[[@B3],[@B4],[@B6]-[@B19]\]. However, little attention has been paid to the stress response of cardiac high risk patients when entering the operating area, during initiation of routine monitoring, and finally during awake venous and arterial canulation. Especially the latter procedure can cause significant discomfort for patients even when performed under local anaesthesia \[[@B20]\]. In several trials a high inter-individual variation of pre-induction norepinephrine levels, heart rate and blood pressure could be noticed \[[@B21]\]. Thus, this observational analysis was performed in patients undergoing coronary aortic bypass surgery (CABG-surgery) in order to identify patients at high risk for increased sympathoadrenergic stress response during the immediate preoperative period using norepinephrine levels and the hemodynamic status as surrogate measures. Methods ======= After Ethics Committee approval was obtained, patients gave their written and informed consent. Eighty-four consecutive male patients undergoing CABG-surgery were enrolled into this observational study. The only exclusion criterion was emergency operation. Due to the observational character of the study, drugs administered preoperatively (including clonidine) were given at the discretion of the anaesthetist performing the preoperative examination. Each patient received an oral premedication with clorazepate 20 mg in the evening before surgery and in the morning of surgery. In about half of the patients, benzodiazepine premedication was combined with clonidine 75--300 μg. Patients were maintained on their regular cardiac and antihypertensive medication up to the day of surgery but all inhibitors of platelet aggregation were discontinued 3--7 day preoperatively. After arrival at the operating theatre an i.v. line was initiated and 500 ml hydroxyethyl-starch (10%, 200000 Dalton) was infused. A 12-lead channel ECG with an automatic ST-segment analysis, oxygen saturation and invasive blood pressure monitoring were connected to the patients. A radial artery catheter was then inserted after local anaesthesia with 1 ml mepivacaine 1%. Heart rate (HR) and mean arterial blood pressure (MAP) were recorded every 15 seconds online using a Laptop computer connected to a Solar 9500 monitor (General Electrics, USA). HR and MAP were multiplied to receive the pressure-rate product (PRP). These variables were used as a measure of the hemodynamic stress response. To determine humoral stress response, an arterial blood sample for the measurement of norepinephrine plasma level was taken immediately after placing the radial artery catheter (this measurement was the main outcome of the study) and 5, 15, and 60 minutes after endotracheal intubation was performed (these measurements were used in an additional explorative analysis, see figure [1](#F1){ref-type="fig"}). 10 ml plastic lithium-heparin tubes were used for this purpose. Specimens were placed on ice immediately after sampling, spun in a centrifuge for 20 minutes and plasma was separated and stored at -70°C pending analysis. Plasma norepinephrine levels were determined by high-performance liquid chromatography (HPLC) with electrochemical detection (Millipore, Billerica, Mass. USA). The lower limit of detection for norepinephrine was 0.018--0.024 nmol·l^-1^and the same-day coefficient of variation for norepinephrine measurements determined by repeated measures of a standardized probe was 3%. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Norepinephrine levels in patients with and without clonidine premedication. ::: ![](1471-2253-4-7-1) ::: General anaesthesia was standardized. After administration of midazolam 0.05 mg·kg^-1^and a three minute period of preoxygenation, anaesthesia was induced using a continuous infusion of propofol (10 mg·kg^-1.^h^-1^) and sufentanil (10 μg.kg^-1.^h^-1^). After loss of consciousness propofol was reduced to 3 mg·kg^-1.^h^-1^and sufentanil to 1.5 μg·kg^-1.^h^-1^. Endotracheal intubation was performed after administration of pancuronium bromide 0.1 mg·kg^-1^. To allow comprehensive analysis of potential factors associated with a reduced stress response, the following data were recorded prospectively: • Age • Bodyweight • Height • Body mass index (BMI) • Clorazepate dose per kilogram bodyweight • Clonidine (yes -- no) • Clonidine per kilogram bodyweight • Time from morning premedication until observational period • Inhibitors of the angiotensine converting enzyme system (ace-inhibitors) • Beta-blocking drugs • Calcium antagonists • Angiotensin-2 receptor inhibitors • Left ventricular ejection fraction (EF) • Number of affected vessels Statistical analysis -------------------- A power analysis had revealed that 80 patients provide a power of more than 95% to detect an R^2^of 0.3 and higher, attributed to 14 independent variables using an F-test with a significance level of 0.05. All potential relevant factors were subjected to a stepwise linear regression analysis using a backward technique. In each step the least significant factor was eliminated when F was lower than 3.96. The quality of the regression model was judged using the Durbin-Watson statistic (a value between 0 and 4 indicating the amount of autocorrelation within the model with an optimum of 2.0), and by checking if the standardized residuals are normally distributed. All calculations were performed using SPSS 11.0 for Windows. All continuous data are presented as mean and standard deviation when normally distributed and as median (25^th^--75^th^percentile) when normal distribution had to be rejected using the Kolmogorov-Smirnov-test. Results ======= Stepwise regression analysis revealed that the single administration of low-dose clonidine (mean dose 1.75 μg·kg^-1^) on the morning of surgery was the only significant predictor (p = 0.004) of the high variation in preoperative norepinephrine plasma levels. This intervention decreased norepinephrine levels by 40% compared to no clonidine administration (from 1.26 to 0.75 nmol·l^-1^). In this analysis, the dichotomous variable (clonidine administration: yes-no) was a better predictor for the norepinephrine levels than variables including the clonidine dose (absolute dose or dose per body weight), indicating that our data provide no evidence for a strong dose-responsiveness of clonidine in this context. All other of the investigated factors (see methods) were removed from the regression model as not significant. The two factors that were eliminated with a p \< 0.1 during the last but one and during the final step were body mass index (removed in step 11 with a p = 0.064) and age (removed in step 12 with a p = 0.076). Both factors were associated with increased norepinephrine levels. The overall quality of the regression model was excellent. The Durbin-Watson coefficient was 2.04 (very near to the optimum of 2.0) and the standardized residuals were normally distributed. For mean arterial blood pressure, heart rate and the calculated pressure rate product, however, preoperative clonidine administration was not an influencing factor. For the mean arterial pressure (MAP), a higher ejection fraction (EF) was a statistically significant predictor (p = 0.024). Each 10% increase of EF was associated with a 2.7 mmHg higher MAP. Administration of an ace-inhibitor was the second predictor in the final model of MAP (p = 0.03). These patients had a 7.5 mmHg lower MAP than patients without ace-inhibitors. For heart rate (HR) there were three significant predictors that remained in the model. Administration of beta-blockers and ace-inhibitors were both associated with a decreased HR (p = 0.004). Each of them decreased HR between 6--7 beats per minute (bpm). Additionally, a higher BMI was associated with a 1.3 bpm higher HR per kg·m^-2^(p = 0.001). Since the PRP is the product of HR and MAP, it is not surprising that similar variables contributed to its prediction. These were the administration of beta-blockers (p = 0.017) and ace-inhibitors (p = 0.004), each of them reducing the PRP, whereas the EF was associated with an increase in PRP (p = 0.014). No patient had signs of cardiac ischemia on arrival at the operating theatre until induction of anaesthesia (defined as ST-T change \> 0.1 mV in any ECG lead). There were no major adverse events during the entire induction period and surgery. Discussion ========== The α~2~-adrenergic receptor agonist clonidine acts by decreasing central sympathetic nervous system activity in all hyperadrenergic situations. In addition to its sedative, anxiolytic, analgesic and antihypertensive properties \[[@B6],[@B22]\] it has shown to improve congestive heart failure, to optimize the myocardial oxygen supply / demand ratio in ischemic heart disease \[[@B23],[@B24]\] and to reduce attacks of angina pectoris \[[@B25],[@B26]\]. In many investigations attention has been drawn to the stressful stimulus of endotracheal intubation \[[@B3],[@B4],[@B6]-[@B19]\], as it has been shown that even short-lasting sympathetic cardiovascular stimulation may have detrimental effects on the coronary circulation of patients with coronary artery disease (CAD), with higher rates of morbidity and lethality \[[@B4],[@B5]\], However, little emphasis has been paid to the preoperative period where patients may be stressed by or because of the upcoming procedure. Furthermore, transfer to the operating theatre, initiation of routine monitoring, and venous and arterial canulations are stressors for the patients. In this context it is certainly a drawback of our study that we did not record the level of preoperative sedation or anxiolysis using clinical measurements on appropriate scales. Instead of this, only a rough judgment was made (awake versus asleep but rousable) that did not allow a valid analysis of the data. Previous data, however, have shown potent anxiolytic and sedative properties of the drug \[[@B27]\]. Thus, it was the major aim of this observational study to identify factors that might contribute either to increased humoral stress or that might help to attenuate this response. Our results show, that a single application of low dose oral clonidine was the only factor that was associated with significantly decreased norepinephrine levels on arrival at the pre-induction area. The question that arises from this observation is, if this is simply an association (or statistically spoken a co-linearity between other protective factors) or if clonidine premedication is the cause for lower norepinephrine levels. In our opinion the latter is the case. Firstly, there were no differences considering any other variables between those patients who had received clonidine and those who had not (see table [1](#T1){ref-type="table"}). Thus, it is unlikely that other factors were responsible for the reduced norepinephrine levels. Secondly, there is good evidence from the literature that clonidine is a powerful drug that attenuates stress response of various causes \[[@B3],[@B4],[@B6]-[@B19]\]. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Patients\' demographic data and preoperative condition. Data are presented for all 84 patients that were included in the study, and separately for those patients receiving clonidine and those without oral clonidine premedication. Values are expressed as mean ± standard deviation, median (25^th^--75^th^percentile), or n = (%). ::: -------------------------------------------------------------------------------------------------------------------------------------------------------------- All patients Patients with clonidine morning premedication Patients without clonidine morning premedication ------------------------------------------ ---------------- ----------------------------------------------- -------------------------------------------------- Age (years) 66 ± 9 65 ± 9 67 ± 8 Bodyweight (kg) 82 ± 10 82 ± 11 82 ± 10 Height (cm) 173 ± 6 173 ± 6 173 ± 5 BMI (kg · m^-2^) 27.3 ± 3.1 27.3 ± 3.2 27.4 ± 3.0 EF (%) 62 ± 14 61 ± 13 63 ± 16 Affected vessels (n = / %) n = 1 3 (4%) 0 (0%) 3 (7%) n = 2 16 (19%) 9 (21%) 7 (17%) n = 3 64 (76%) 33 (79%) 31 (74%) n = 4 1 (1%) 0 (0%) 1 (2%) Pre-treated with (n= / %) ace-inhibitors 43 (51%) 17 (40%) 26 (62%) Beta-blockers 55 (65%) 26 (62%) 29 (69%) Calcium-antagonists 12 (14%) 5 (12%) 7 (17%) Clonidine dose (n= / %) 75 μg 8 (19%) 150 μg n/a 33 (79%) n/a 300 μg 1 (2%) Time from morning premedication\ 1.0 (0.5--4.5) 2.5 (0.5--5.0) 1.0 (0.5--4.5) until observational period (hours) Heart rate \[bpm\] 66 ± 11 66 ± 10 66 ± 12 Mean arterial blood pressure \[mmHg\] 102 ± 16 100 ± 15 104 ± 17 Pressure rate product \[mmHg·bpm\] 6750 ± 1640 6660 ± 1600 6850 ± 1690 Plasma norepinephrine level (nmol·l^-1^) 1.00 ± 0.82 0.75 ± 0.48 1.26 ± 1.00 -------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: However, it is interesting to notice that the mean dose administered to our patients (1.75 μg·kg^-1^) was low compared to all other trials. Data concerning the appropriate dose of clonidine to attenuate the stress response to intubation vary considerably between 0.625 and 10 μg·kg^-1^. For example, one trial demonstrated that clonidine 0.625 and 1.25 μg·kg^-1^i.v. were sufficient to attenuate pressure response to laryngoscopy and intubation \[[@B28]\], whereas in another one \[[@B19]\] evaluating the dose-response effects to laryngoscopy and intubation, 2 μg·kg^-1^clonidine i.v. was equally effective as placebo, and only 4 and 6 μg·kg^-1^significantly attenuated hemodynamic and adrenergic reactions in an equal manner. It could also be shown that 4 or 6 μg·kg^-1^were necessary to reduce norepinephrine levels beforeinduction of anaesthesia, however 2 μg·kg^-1^where not sufficient in this setting \[[@B19]\]. In our trial as well as in all other studies with even much higher doses, clonidine was well tolerated and did not produce any adverse hemodynamic effects. In our analysis there was no strong evidence for a dose responsiveness of orally administered clonidine. First, in the regression model catecholamine levels could better be predicted by the dichotomous variable and second, there was only a weak correlation between the weight adjusted clonidine dose on the one hand and norepinephrine levels on the other hand (Pearson\'s correlation coefficient was -0.31, Spearman\'s rank correlation coefficient was -0.30). Furthermore, a post-hoc comparison between the patients receiving either 75 or 150μg clonidine did not show relevant differences (p = 0.91 using the Mann-Whitney U-test). Higher age and higher body mass index showed a non-significant tendency to increase the catecholamine concentration. No other of the investigated factors (body weight, height, time from morning premedication until observational period, benzodiazepine dose per kilogram bodyweight, ace-inhibitors, beta-blocking drugs, calcium antagonists, EF, number of affected vessels) had statistically significant impact on norepinephrine levels. An explorative post-hoc analysis of the impact of clonidine premedication (none versus any dose) and clonidine dose on norepinephrine levels during the entire induction period proves the results of the main analysis. There was a pronounced reduction of norepinephrine plasma levels after induction of general anaesthesia with lower values in the clonidine-group. However, a statistically significant interaction term (p = 0.012) suggests that the fall of norepinephrine levels are more marked in the untreated group and thus mainly caused by induction of general anaesthesia rather than effects of clonidine (figure [1](#F1){ref-type="fig"}). Conclusion ========== This observational trial demonstrates that patients undergoing coronary artery bypass graft surgery have a great variation of norepinephrine levels when entering the operating theatre. We could identify oral clonidine premedication as the only predictor for increased humoral stress response. There was no strong evidence for a dose dependency, indicating that even small doses, like 75--150 μg attenuate the humoral stress response before coronary artery bypass graft surgery. Clonidine did not have a negative impact on hemodynamic parameters. Competing interests =================== None declared. Authors\' contributions ======================= AMM processed the data and wrote the manuscript. GG conceived the study, collected the clinical data and participated in its design. US collected the clinical data. MK designed the study and collected the clinical data. HAA performed the laboratory investigations. HW participated in the conception of the study. LHJE designed the study, performed the statistical analysis and extensively revised 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/1471-2253/4/7/prepub>	PubMed Central	2024-06-05T03:55:48.329997	2004-9-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521687/", "journal": "BMC Anesthesiol. 2004 Sep 23; 4:7", "authors": [ { "first": "Astrid M", "last": "Morin" }, { "first": "Götz", "last": "Geldner" }, { "first": "Udo", "last": "Schwarz" }, { "first": "Martin", "last": "Kahl" }, { "first": "Hans A", "last": "Adams" }, { "first": "Hinnerk", "last": "Wulf" }, { "first": "Leopold HJ", "last": "Eberhart" } ] }
PMC521688	Background ========== Consumers of research (clinicians, researchers, educators, administrators) frequently use standard critical appraisal tools to evaluate the quality and utility of published research reports \[[@B1]\]. Critical appraisal tools provide analytical evaluations of the quality of the study, in particular the methods applied to minimise biases in a research project \[[@B2]\]. As these factors potentially influence study results, and the way that the study findings are interpreted, this information is vital for consumers of research to ascertain whether the results of the study can be believed, and transferred appropriately into other environments, such as policy, further research studies, education or clinical practice. Hence, choosing an appropriate critical appraisal tool is an important component of evidence-based practice. Although the importance of critical appraisal tools has been acknowledged \[[@B1],[@B3]-[@B5]\] there appears to be no consensus regarding the \'gold standard\' tool for any medical evidence. In addition, it seems that consumers of research are faced with a large number of critical appraisal tools from which to choose. This is evidenced by the recent report by the Agency for Health Research Quality in which 93 critical appraisal tools for quantitative studies were identified \[[@B6]\]. Such choice may pose problems for research consumers, as dissimilar findings may well be the result when different critical appraisal tools are used to evaluate the same research report \[[@B6]\]. Critical appraisal tools can be broadly classified into those that are research design-specific and those that are generic. Design-specific tools contain items that address methodological issues that are unique to the research design \[[@B5],[@B7]\]. This precludes comparison however of the quality of different study designs \[[@B8]\]. To attempt to overcome this limitation, generic critical appraisal tools have been developed, in an attempt to enhance the ability of research consumers to synthesise evidence from a range of quantitative and or qualitative study designs (for instance \[[@B9]\]). There is no evidence that generic critical appraisal tools and design-specific tools provide a comparative evaluation of research designs. Moreover, there appears to be little consensus regarding the most appropriate items that should be contained within any critical appraisal tool. This paper is concerned primarily with critical appraisal tools that address the unique properties of allied health care and research \[[@B10]\]. This approach was taken because of the unique nature of allied health contacts with patients, and because evidence-based practice is an emerging area in allied health \[[@B10]\]. The availability of so many critical appraisal tools (for instance \[[@B6]\]) may well prove daunting for allied health practitioners who are learning to critically appraise research in their area of interest. For the purposes of this evaluation, allied health is defined as encompassing \"\...all occasions of service to non admitted patients where services are provided at units/clinics providing treatment/counseling to patients. These include units primarily concerned with physiotherapy, speech therapy, family panning, dietary advice, optometry occupational therapy\...\" \[[@B11]\]. The unique nature of allied health practice needs to be considered in allied health research. Allied health research thus differs from most medical research, with respect to: • the paradigm underpinning comprehensive and clinically-reasoned descriptions of diagnosis (including validity and reliability). An example of this is in research into low back pain, where instead of diagnosis being made on location and chronicity of pain (as is common) \[[@B12]\], it would be made on the spinal structure and the nature of the dysfunction underpinning the symptoms, which is arrived at by a staged and replicable clinical reasoning process \[[@B10],[@B13]\]. • the frequent use of multiple interventions within the one contact with the patient (an occasion of service), each of which requires appropriate description in terms of relationship to the diagnosis, nature, intensity, frequency, type of instruction provided to the patient, and the order in which the interventions were applied \[[@B13]\] • the timeframe and frequency of contact with the patient (as many allied health disciplines treat patients in episodes of care that contain multiple occasions of service, and which can span many weeks, or even years in the case of chronic problems \[[@B14]\]) • measures of outcome, including appropriate methods and timeframes of measuring change in impairment, function, disability and handicap that address the needs of different stakeholders (patients, therapists, funders etc) \[[@B10],[@B12],[@B13]\]. Methods ======= Search strategy --------------- In supplementary data \[see [additional file 1](#S1){ref-type="supplementary-material"}\]. Data organization and extraction -------------------------------- Two independent researchers (PK, NMW) participated in all aspects of this review, and they compared and discussed their findings with respect to inclusion of critical appraisal tools, their intent, components, data extraction and item classification, construction and psychometric properties. Disagreements were resolved by discussion with a third member of the team (KG). Data extraction consisted of a four-staged process. First, identical replica critical appraisal tools were identified and removed prior to analysis. The remaining critical appraisal tools were then classified according to the study design for which they were intended to be used \[[@B1],[@B2]\]. The scientific manner in which the tools had been constructed was classified as whether an empirical research approach has been used, and if so, which type of research had been undertaken. Finally, the items contained in each critical appraisal tool were extracted and classified into one of eleven groups, which were based on the criteria described by Clarke and Oxman \[[@B4]\] as: • [Study aims and justification]{.underline} • [Methodology used]{.underline}, which encompassed method of identification of relevant studies and adherence to study protocol; • [Sample selection]{.underline}, which ranged from inclusion and exclusion criteria, to homogeneity of groups; • Method of [randomization]{.underline} and allocation blinding; • [Attrition]{.underline}: response and drop out rates; • [Blinding]{.underline} of the clinician, assessor, patient and statistician as well as the method of blinding; • [Outcome measure]{.underline} characteristics; • [Intervention]{.underline} or exposure details; • Method of [data analyses]{.underline}; • Potential sources of [bias]{.underline}; and • Issues of [external validity]{.underline}, which ranged from application of evidence to other settings to the relationship between benefits, cost and harm. An additional group, \"[miscellaneous]{.underline}\", was used to describe items that could not be classified into any of the groups listed above. Data synthesis -------------- Data was synthesized using MS Excel spread sheets as well as narrative format by describing the number of critical appraisal tools per study design and the type of items they contained. Descriptions were made of the method by which the overall quality of the study was determined, evidence regarding the psychometric properties of the tools (validity and reliability) and whether guidelines were provided for use of the critical appraisal tool. Results ======= One hundred and ninety-three research reports that potentially provided a description of a critical appraisal tool (or process) were identified from the search strategy. Fifty-six of these papers were unavailable for review due to outdated Internet links, or inability to source the relevant journal through Australian university and Government library databases. Of the 127 papers retrieved, 19 were excluded from this review, as they did not provide a description of the critical appraisal tool used, or were published in languages other than English. As a result, 108 papers were reviewed, which yielded 121 different critical appraisal tools \[[@B1]-[@B5],[@B7],[@B9],[@B15]-[@B102],[@B116]\]. Empirical basis for tool construction ------------------------------------- We identified 14 instruments (12% all tools) which were reported as having been constructed using a specified empirical approach \[[@B20],[@B29],[@B30],[@B32],[@B35],[@B40],[@B49],[@B51],[@B70]-[@B72],[@B79],[@B103],[@B116]\]. The empirical research reflected descriptive and/or qualitative approaches, these being critical review of existing tools \[[@B40],[@B72]\], Delphi techniques to identify then refine data items \[[@B32],[@B51],[@B71]\], questionnaires and other forms of written surveys to identify and refine data items \[[@B70],[@B79],[@B103]\], facilitated structured consensus meetings \[[@B20],[@B29],[@B30],[@B35],[@B40],[@B49],[@B70],[@B72],[@B79],[@B116]\], and pilot validation testing \[[@B20],[@B40],[@B72],[@B103],[@B116]\]. In all the studies which reported developing critical appraisal tools using a consensus approach, a range of stakeholder input was sought, reflecting researchers and clinicians in a range of health disciplines, students, educators and consumers. There were a further 31 papers which cited other studies as the source of the tool used in the review, but which provided no information on why individual items had been chosen, or whether (or how) they had been modified. Moreover, for 21 of these tools, the cited sources of the critical appraisal tool did not report the empirical basis on which the tool had been constructed. Critical appraisal tools per study design ----------------------------------------- Seventy-eight percent (N = 94) of the critical appraisal tools were developed for use on primary research \[[@B1]-[@B5],[@B7],[@B9],[@B18],[@B19],[@B25]-[@B27],[@B34],[@B37]-[@B41]\], while the remainder (N = 26) were for secondary research (systematic reviews and meta-analyses) \[[@B2]-[@B5],[@B15]-[@B36],[@B116]\]. Eighty-seven percent (N = 104) of all critical appraisal tools were design-specific \[[@B2]-[@B5],[@B7],[@B9],[@B15]-[@B90]\], with over one third (N = 45) developed for experimental studies (randomized controlled trials, clinical trials) \[[@B2]-[@B4],[@B25]-[@B27],[@B34],[@B37]-[@B73]\]. Sixteen critical appraisal tools were generic. Of these, six were developed for use on both experimental and observational studies \[[@B9],[@B91]-[@B95]\], whereas 11 were purported to be useful for any qualitative and quantitative research design \[[@B1],[@B18],[@B41],[@B96]-[@B102],[@B116]\] (see Figure [1](#F1){ref-type="fig"}, Table [1](#T1){ref-type="table"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Number of critical appraisal tools per study design \[1,2\] ::: ![](1471-2288-4-22-1) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Summary of tools sourced in this review. ::: Research design focus of critical appraisal tools Critical appraisal tools with summary scores ------------------------------------------------------- ---------------------------------------------------- ---------------------------------------------- -------------------------------------------- Secondary studies Systematic reviews/meta-analyses \[2-5,15-36,116\] All study designs \[1,18,41,96-102,116\] Summary score \[18,41,96,97,116\] Primary studies Experimental studies \[2-4,19,25-27,34,37-73\] No summary score \[1,98-102\] Diagnostic studies \[19,74-79\] Experimental studies Summary score \[19,37-59\] Observational studies \[2,3,7,19,25,66,72,80-86\] No summary score \[2-4,25,27,28,34,60-73\] Qualitative studies \[9,26,66,87-90\] Diagnostic studies Summary score \[16,74-77\] Experimental & Observational studies \[9,91-102\] No summary score \[78,79\] Qualitative studies Summary score \[87\] No summary score \[9,26,66,88-90\] Experimental and observational studies Summary score \[91-93\] No summary score \[9,94,95\] ::: Critical appraisal items ------------------------ One thousand, four hundred and seventy five items were extracted from these critical appraisal tools. After grouping like items together, 173 different item types were identified, with the most frequently reported items being focused towards assessing the external validity of the study (N = 35) and method of data analyses (N = 28) (Table [2](#T2){ref-type="table"}). The most frequently reported items across all critical appraisal tools were: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### The type and number of component items contained in critical appraisal tools per study design. ::: Type of items Design-specific critical appraisal tool components Generic critical appraisal tool components Total --------------------------------- -------------------------------------------------------- ------------------------------------------------ ----------- ----- ---- ---- ----- ------ Study aims and justification 35 27 5 18 17 4 11 117 Methodology used 38 1 0 0 0 0 1 40 Sample selection 30 62 12 37 10 10 14 175 Randomization 2 65 1 5 0 6 5 84 Attrition 4 59 3 23 0 8 8 105 Blinding 1 77 5 8 0 5 7 103 Outcome measure characteristics 41 46 3 33 2 9 19 153 Intervention 7 42 3 13 0 5 12 82 Data analyses 83 91 14 54 12 14 27 295 Bias 24 14 2 5 0 3 6 54 External validity 72 50 12 30 27 9 27 227 Miscellaneous 11 12 7 5 7 2 6 50 Total 348 546 67 331 75 75 143 1485 ::: • Eligibility criteria (inclusion/exclusion criteria) (N = 63) • Appropriate statistical analyses (N = 47) • Random allocation of subjects (N = 43) • Consideration of outcome measures used (N = 43) • Sample size justification/power calculations (N = 39) • Study design reported (N = 36) • Assessor blinding (N = 36) Design-specific critical appraisal tools ---------------------------------------- ### Systematic reviews Eighty-seven different items were extracted from the 26 critical appraisal tools, which were designed to evaluate the quality of systematic reviews. These critical appraisal tools frequently contained items regarding data analyses and issues of external validity (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The type and number of guidelines accompanying critical appraisal tools per study design ::: Type of critical appraisal tool Type of guideline Total number of critical appraisal tools -------------------------------------- ----------------------- ---------------------------------------------- -------- ----------------------- -------- --------- Systematic reviews 9 \[2,4,15,20,25,28,29,331,36,116\] 3 \[16,26,27\] 12 26 Experimental studies 10 \[2,4,25,37,41,50,64-66,69\] 6 \[26,40,49,51,57,59\] 16 45 Diagnostic studies 3 \[74,75,76\] 1 \[79\] 4 7 Observational studies 9 \[2,25,66,80,84-87\] 1 \[83\] 10 19 Qualitative studies 4 \[9,87,89,90\] 1 \[26\] 5 7 Experimental & Observational studies 2 \[9,95\] 1 \[91\] 3 6 All study designs 1 \[100\] 1 \[102\] 2 10 Total 38 14 52 120 ::: Items assessing data analyses were focused to the methods used to summarize the results, assessment of sensitivity of results and whether heterogeneity was considered, whereas the nature of reporting of the main results, interpretation of them and their generalizability were frequently used to assess the external validity of the study findings. Moreover, systematic review critical appraisal tools tended to contain items such as identification of relevant studies, search strategy used, number of studies included and protocol adherence, that would not be relevant for other study designs. Blinding and randomisation procedures were rarely included in these critical appraisal tools. ### Experimental studies One hundred and twenty thirteen different items were extracted from the 45 experimental critical appraisal tools. These items most frequently assessed aspects of data analyses and blinding (Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). Data analyses items were focused on whether appropriate statistical analysis was performed, whether a sample size justification or power calculation was provided and whether side effects of the intervention were recorded and analysed. Blinding was focused on whether the participant, clinician and assessor were blinded to the intervention. ### Diagnostic studies Forty-seven different items were extracted from the seven diagnostic critical appraisal tools. These items frequently addressed issues involving data analyses, external validity of results and sample selection that were specific to diagnostic studies (whether the diagnostic criteria were defined, definition of the \"gold\" standard, the calculation of sensitivity and specificity) (Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). ### Observational studies Seventy-four different items were extracted from the 19 critical appraisal tools for observational studies. These items primarily focused on aspects of data analyses (see Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}, such as whether confounders were considered in the analysis, whether a sample size justification or power calculation was provided and whether appropriate statistical analyses were preformed. ### Qualitative studies Thirty-six different items were extracted from the seven qualitative study critical appraisal tools. The majority of these items assessed issues regarding external validity, methods of data analyses and the aims and justification of the study (Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). Specifically, items were focused to whether the study question was clearly stated, whether data analyses were clearly described and appropriate, and application of the study findings to the clinical setting. Qualitative critical appraisal tools did not contain items regarding sample selection, randomization, blinding, intervention or bias, perhaps because these issues are not relevant to the qualitative paradigm. ### Generic critical appraisal tools #### Experimental and observational studies Forty-two different items were extracted from the six critical appraisal tools that could be used to evaluate experimental and observational studies. These tools most frequently contained items that addressed aspects of sample selection (such as inclusion/exclusion criteria of participants, homogeneity of participants at baseline) and data analyses (such as whether appropriate statistical analyses were performed, whether a justification of the sample size or power calculation were provided). #### All study designs Seventy-eight different items were contained in the ten critical appraisal tools that could be used for all study designs (quantitative and qualitative). The majority of these items focused on whether appropriate data analyses were undertaken (such as whether confounders were considered in the analysis, whether a sample size justification or power calculation was provided and whether appropriate statistical analyses were preformed) and external validity issues (generalization of results to the population, value of the research findings) (see Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). Allied health critical appraisal tools -------------------------------------- We found no critical appraisal instrument specific to allied health research, despite finding at least seven critical appraisal instruments associated with allied health topics (mostly physiotherapy management of orthopedic conditions) \[[@B37],[@B39],[@B52],[@B58],[@B59],[@B65]\]. One critical appraisal development group proposed two instruments \[[@B9]\], specific to quantitative and qualitative research respectively. The core elements of allied health research quality (specific diagnosis criteria, intervention descriptions, nature of patient contact and appropriate outcome measures) were not addressed in any one tool sourced for this evaluation. We identified 152 different ways of considering quality reporting of outcome measures in the 121 critical appraisal tools, and 81 ways of considering description of interventions. Very few tools which were not specifically targeted to diagnostic studies (less than 10% of the remaining tools) addressed diagnostic criteria. The critical appraisal instrument that seemed most related to allied health research quality \[[@B39]\] sought comprehensive evaluation of elements of intervention and outcome, however this instrument was relevant only to physiotherapeutic orthopedic experimental research. Overall study quality --------------------- Forty-nine percent (N = 58) of critical appraisal tools summarised the results of the quality appraisal into a single numeric summary score \[[@B5],[@B7],[@B15]-[@B25],[@B37]-[@B59],[@B74]-[@B77],[@B80]-[@B83],[@B87],[@B91]-[@B93],[@B96],[@B97]\] (Figure [2](#F2){ref-type="fig"}). This was achieved by one of two methods: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Number of critical appraisal tools with, and without, summary quality scores ::: ![](1471-2288-4-22-2) ::: • An equal weighting system, where one point was allocated to each item fulfilled; or • A weighted system, where fulfilled items were allocated various points depending on their perceived importance. However, there was no justification provided for any of the scoring systems used. In the remaining critical appraisal tools (N = 62), a single numerical summary score was not provided \[[@B1]-[@B4],[@B9],[@B25]-[@B36],[@B60]-[@B73],[@B78],[@B79],[@B84]-[@B90],[@B94],[@B95],[@B98]-[@B102]\]. This left the research consumer to summarize the results of the appraisal in a narrative manner, without the assistance of a standard approach. Psychometric properties of critical appraisal tools --------------------------------------------------- Few critical appraisal tools had documented evidence of their validity and reliability. Face validity was established in nine critical appraisal tools, seven of which were developed for use on experimental studies \[[@B38],[@B40],[@B45],[@B49],[@B51],[@B63],[@B70]\] and two for systematic reviews \[[@B32],[@B103]\]. Intra-rater reliability was established for only one critical appraisal tool as part of its empirical development process \[[@B40]\], whereas inter-rater reliability was reported for two systematic review tools \[[@B20],[@B36]\] (for one of these as part of the developmental process \[[@B20]\]) and seven experimental critical appraisal tools \[[@B38],[@B40],[@B45],[@B51],[@B55],[@B56],[@B63]\] (for two of these as part of the developmental process \[[@B40],[@B51]\]). Critical appraisal tool guidelines ---------------------------------- Forty-three percent (N = 52) of critical appraisal tools had guidelines that informed the user of the interpretation of each item contained within them (Table [2](#T2){ref-type="table"}). These guidelines were most frequently in the form of a handbook or published paper (N = 31) \[[@B2],[@B4],[@B9],[@B15],[@B20],[@B25],[@B28],[@B29],[@B31],[@B36],[@B37],[@B41],[@B50],[@B64]-[@B67],[@B69],[@B80],[@B84]-[@B87],[@B89],[@B90],[@B95],[@B100],[@B116]\], whereas in 14 critical appraisal tools explanations accompanied each item \[[@B16],[@B26],[@B27],[@B40],[@B49],[@B51],[@B57],[@B59],[@B79],[@B83],[@B91],[@B102]\]. Discussion ========== Our search strategy identified a large number of published critical appraisal tools that are currently available to critically appraise research reports. There was a distinct lack of information on tool development processes in most cases. Many of the tools were reported to be modifications of other published tools, or reflected specialty concerns in specific clinical or research areas, without attempts to justify inclusion criteria. Less than 10 of these tools were relevant to evaluation of the quality of allied health research, and none of these were based on an empirical research approach. We are concerned that although our search was systematic and extensive \[[@B104],[@B105]\], our broad key words and our lack of ready access to 29% of potentially useful papers (N = 56) potentially constrained us from identifying all published critical appraisal tools. However, consumers of research seeking critical appraisal instruments are not likely to seek instruments from outdated Internet links and unobtainable journals, thus we believe that we identified the most readily available instruments. Thus, despite the limitations on sourcing all possible tools, we believe that this paper presents a useful synthesis of the readily available critical appraisal tools. The majority of the critical appraisal tools were developed for a specific research design (87%), with most designed for use on experimental studies (38% of all critical appraisal tools sourced). This finding is not surprising as, according to the medical model, experimental studies sit at or near the top of the hierarchy of evidence \[[@B2],[@B8]\]. In recent years, allied health researchers have strived to apply the medical model of research to their own discipline by conducting experimental research, often by using the randomized controlled trial design \[[@B106]\]. This trend may be the reason for the development of experimental critical appraisal tools reported in allied health-specific research topics \[[@B37],[@B39],[@B52],[@B58],[@B59],[@B65]\]. We also found a considerable number of critical appraisal tools for systematic reviews (N = 26), which reflects the trend to synthesize research evidence to make it relevant for clinicians \[[@B105],[@B107]\]. Systematic review critical appraisal tools contained unique items (such as identification of relevant studies, search strategy used, number of studies included, protocol adherence) compared with tools used for primary studies, a reflection of the secondary nature of data synthesis and analysis. In contrast, we identified very few qualitative study critical appraisal tools, despite the presence of many journal-specific guidelines that outline important methodological aspects required in a manuscript submitted for publication \[[@B108]-[@B110]\]. This finding may reflect the more traditional, quantitative focus of allied health research \[[@B111]\]. Alternatively, qualitative researchers may view the robustness of their research findings in different terms compared with quantitative researchers \[[@B112],[@B113]\]. Hence the use of critical appraisal tools may be less appropriate for the qualitative paradigm. This requires further consideration. Of the small number of generic critical appraisal tools, we found few that could be usefully applied (to any health research, and specifically to the allied health literature), because of the generalist nature of their items, variable interpretation (and applicability) of items across research designs, and/or lack of summary scores. Whilst these types of tools potentially facilitate the synthesis of evidence across allied health research designs for clinicians, their lack of specificity in asking the \'hard\' questions about research quality related to research design also potentially precludes their adoption for allied health evidence-based practice. At present, the gold standard study design when synthesizing evidence is the randomized controlled trial \[[@B4]\], which underpins our finding that experimental critical appraisal tools predominated in the allied health literature \[[@B37],[@B39],[@B52],[@B58],[@B59],[@B65]\]. However, as more systematic literature reviews are undertaken on allied health topics, it may become more accepted that evidence in the form of other research design types requires acknowledgement, evaluation and synthesis. This may result in the development of more appropriate and clinically useful allied health critical appraisal tools. A major finding of our study was the volume and variation in available critical appraisal tools. We found no gold standard critical appraisal tool for any type of study design. Therefore, consumers of research are faced with frustrating decisions when attempting to select the most appropriate tool for their needs. Variable quality evaluations may be produced when different critical appraisal tools are used on the same literature \[[@B6]\]. Thus, interpretation of critical analysis must be carefully considered in light of the critical appraisal tool used. The variability in the content of critical appraisal tools could be accounted for by the lack of any empirical basis of tool construction, established validity of item construction, and the lack of a gold standard against which to compare new critical tools. As such, consumers of research cannot be certain that the content of published critical appraisal tools reflect the most important aspects of the quality of studies that they assess \[[@B114]\]. Moreover, there was little evidence of intra- or inter-rater reliability of the critical appraisal tools. Coupled with the lack of protocols for use, this may mean that critical appraisers could interpret instrument items in different ways over repeated occasions of use. This may produce variable results \[123\]. Conclusions =========== Based on the findings of this evaluation, we recommend that consumers of research should carefully select critical appraisal tools for their needs. The selected tools should have published evidence of the empirical basis for their construction, validity of items and reliability of interpretation, as well as guidelines for use, so that the tools can be applied and interpreted in a standardized manner. Our findings highlight the need for consensus to be reached regarding the important and core items for critical appraisal tools that will produce a more standardized environment for critical appraisal of research evidence. As a consequence, allied health research will specifically benefit from having critical appraisal tools that reflect best practice research approaches which embed specific research requirements of allied health disciplines. Competing interests =================== No competing interests. Authors\' contributions ======================= PK Sourced critical appraisal tools Categorized the content and psychometric properties of critical appraisal tools AEB Synthesis of findings Drafted manuscript NMW Sourced critical appraisal tools Categorized the content and psychometric properties of critical appraisal tools VSK Sourced critical appraisal tools Categorized the content and psychometric properties of critical appraisal tools KAG Study conception and design Assisted with critiquing critical appraisal tools and categorization of the content and psychometric properties of critical appraisal tools Drafted and reviewed manuscript Addressed reviewer\'s comments and re-submitted the article Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2288/4/22/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Search Strategy. ::: ::: {.caption} ###### Click here for file :::	PubMed Central	2024-06-05T03:55:48.332149	2004-9-16	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521688/", "journal": "BMC Med Res Methodol. 2004 Sep 16; 4:22", "authors": [ { "first": "Persis", "last": "Katrak" }, { "first": "Andrea E", "last": "Bialocerkowski" }, { "first": "Nicola", "last": "Massy-Westropp" }, { "first": "VS Saravana", "last": "Kumar" }, { "first": "Karen A", "last": "Grimmer" } ] }
PMC521689	Background ========== Human airway surface is covered by a thin layer of liquid (airway surface liquid \[ASL\]) that contains many antimicrobial substances including lysozyme, lactoferrin, human β defensins, and the cathelicidin LL-37 \[[@B1]-[@B4]\]. The antibacterial activity of most of these innate immune mediators is salt-sensitive; an increase in salt concentration inhibits their activity \[[@B5]\]. An equally interesting feature of these antimicrobial factors is that their activity is increased by low ionic strength \[[@B6]-[@B9]\]. Lowering the ASL salt concentration might therefore increase the efficacy of the innate immune system and thereby decrease or prevent airway infections. The airway epithelium is water-permeable \[[@B10]\]. When large volumes of ionic, isotonic liquid are placed on the apical surface, active salt and liquid absorption occurs \[[@B11],[@B12]\]. If water were added to the airway surface, the salt concentration would quickly return to starting values. Thus, lowering of ASL salt concentration is best accomplished using a nonionic osmolyte with low transepithelial permeability. The osmolyte should not provide a ready carbon source for bacteria, and should be safe in humans. One such promising osmolyte is xylitol, a five-carbon sugar that has low transepithelial permeability, is poorly metabolized by bacteria and can lower the salt concentration of both cystic fibrosis (CF) and non-CF epithelia in vitro\[[@B13]\]. Xylitol is an artificial sweetener that has been successfully used in chewing gums to prevent dental caries \[[@B14],[@B15]\]; it has been used as an oral sugar substitute without significant adverse effects \[[@B16]\]. It has also been used in lozenges and syrup and has been shown to decrease the incidence of acute otitis media by 20--40% \[[@B17]\]; nasal application to normal human subjects was found to decrease colonization with coagulase negative staphylococcus \[[@B13]\]. There are no studies, to our knowledge, examining the effects of inhalation of aerosolized xylitol by experimental animals or humans. Osmotic agents such as hypertonic saline, which is ionic, and nonionic mannitol, dextran and lactose, have been used in human subjects to increase mucus clearance \[[@B18]-[@B23]\]. However, some of these agents can serve as a carbon source for bacteria and can cause bronchospasm due to the tonicity. Nebulization of distilled water has been shown to increase airway resistance significantly in asthmatic subjects leading to subsequent use as a bronchoprovocative agent \[[@B24]-[@B26]\]. Both hypotonic and hypertonic saline solutions can provoke bronchospasm (a 20% drop in Forced Expiratory Volume in 1 second, FEV1) in asthmatic subjects but not in normal volunteers \[[@B26]\]. Furthermore, inhalation of 20% dextrose in the same study produced bronchospasm similar to exposure to water or hypertonic saline raising the possibility that osmolarity of the solution is the important determinant of bronchial reactivity. In subjects with bronchiectasis, inhalation of dry powdered mannitol can increase the clearance of mucus without affecting lung function \[[@B27]\]. However, in a different study on subjects with CF, inhaled mannitol caused a small but significant decline in FEV1 (7.3%, P = 0.004) from baseline immediately after inhalation, which returned to baseline by the end of the study \[[@B28]\]. We hypothesized that aerosolized iso-osmolar xylitol is safe and well-tolerated well by normal subjects. We compared the safety and tolerability of aerosolized xylitol with normal saline, and carried out additional exposure studies using mice. Methods ======= Safety in normal mice --------------------- All experiments were reviewed and approved by the animal care and use committee of the University of Iowa. Except during exposures and evaluation, mice were allowed access to food and water ad libitum. C57bl/6 mice (Jackson Lab, Bar Harbor, MA) underwent baseline methacholine challenge test using a whole-body plethysmograph (Buxco Electronics, Troy, NY) as previously described \[[@B29]\]. Respiratory pattern changes were expressed as enhanced respiratory pause (Penh), which correlates with changes in airway resistance. Airway resistance was expressed as follows: P~enh~= (\[T~e~/0.3 T~r~\] - 1) × \[2P~ef~/3P~if~\], where P~enh~equals enhanced pause, T~e~equals expiratory time (in seconds), T~r~equals relaxation time (in seconds), P~ef~equals peak expiratory flow (in milliliters per second), and P~if~equals peak inspiratory flow (in milliliters per second). Mice (6/group) were exposed to aerosolized saline (0.9 % NaCl) or aerosolized xylitol (5% solution in water, equimolar to the NaCl) for 150 minutes in an exposure chamber; all mice were evaluated for bronchial hyperreactivity to inhaled methacholine (using the Buxco whole body plethysmography system) before and after the exposures; other mice were monitored periodically during exposure by whole body plethysmography. All mice underwent whole lung lavage the next day for cell count and differential. After euthanasia, the trachea was cannulated, and the lungs were lavaged with 3.0 mL of sterile normal saline (0.9% NaCl). The lavage samples were immediately processed for total and differential (with Diff Quick Stain; Baxter Scientific, Miami, FL) cell counts. In a separate group of naïve mice, whole body plethysmography was used to monitor Penh, respiratory rate, and tidal volume periodically during exposure to xylitol and saline for 10, 20, 40, and 80 minutes for a cumulative total dose of 150 minutes. Safety in hypersensitive mice ----------------------------- We repeated the saline and xylitol exposure protocol to 2 more groups of six mice each after they were sensitized to and challenged with an antigen \[[@B30]\]. Mice were sensitized to OVA (10 μg with 1 mg alum, i.p.) on days 0 and 7, then challenged with aerosolized OVA (1% solution, 30 minutes) on days 14 and 16. Filtered air was passed at 6 L/min through an Aero-Tech nebulizer (CIS-US Inc) to generate an aerosol. The size distribution of the aerosol was determined using a particle counter (Aerodynamic Particle Sizer, TSI Incorporated). The aerosol sizes were distributed log normally with a count median aerodynamic diameter of 0.82 microns and geometric standard deviation (GSD) of 1.46 microns. A mean OVA concentration of 3.8 ng/ml was measured in the chamber during the exposures. The mice underwent a baseline methacholine challenge on day 17 and subsequently underwent exposures to saline and xylitol using the same protocol described for the naïve mice. Three mice per group underwent whole lung lavage 24 hours after exposure for cell count and differential. Given the concerns that have been raised about the reliability of airway resistance measurement by Buxco equipment, in a select number of mice we confirmed airway hyperresponsiveness using invasive measurement. Airway responsiveness was measured 24 hours after xylitol exposure in ova-challenged mice and compared to measurements made on naïve mice and ova-challenged mice without any exposure. Mice were anesthetized with Ketamine at 90 mg/kg and Pentobarbital at 50 mg/kg and attached to a small-animal ventilator (Flexivent, SCIREQ). Animals were ventilated at 150 breaths/min. Positive end-expiratory pressure (PEEP) was maintained between 2--3 cmH2O, with the computer setting the tidal volume from the entered weight of each animal. Central airway resistance (R) was measured at baseline and after 10 sec. of nebulized methacholine at doses of 12.5, 25 and 50 mg/ml. Safety in normal volunteers --------------------------- The study was approved by the University of Iowa Institutional Review Board as well as the Food and Drug Administration. Since this is a pilot study and would be the first time xylitol is being used as aerosol, there was no information available on expected complications. Ten subjects aged 18 or greater were studied. Pregnancy or any chronic medical conditions such asthma, atopy, and diabetes were grounds for exclusion. After giving written informed consent subjects underwent a screening spirometry (all subjects demonstrated FEV1 \>85% of predicted). Baseline measurements of serum electrolytes, and serum and urine osmolarity were carried out. Baseline oxygen saturation was measured using a pulse oximeter. A brief questionnaire of respiratory symptoms that was developed using a visual analog scale (VAS) was administered at baseline \[[@B31],[@B32]\]. Human exposures --------------- Subjects received 10 ml of aerosolized saline (generated using a Pari LC Plus nebulizer with Proneb Ultra compressor system, Pari Inc, Monterey, CA) \[[@B33]\]. The particle size of the aerosol was measured using both a 7-stage cascade impactor (Mercer, Inc., Albuquerque, NM) and an Aerosol Monitor (Grimm Technologies, Inc.). The mass median aerodynamic diameter of the aerosol was 1.63 microns with a GSD of 1.71 microns. Mean breathing time for exposures were as follows: Normal saline -- 37 min (range 22--49), 1 ml xylitol -- 4.2 min (range 2--10), 5 ml xylitol -- 22 min (range 15--33), 10 ml xylitol -- 36 min (range 30--49). Thirty minutes after the exposures, subjects completed a follow-up questionnaire, and underwent spirometry and O2 saturation measurement. The procedure was repeated after exposure to 1, 5, and 10 ml of 5% xylitol (Danisco Cultor, USA). Xylitol was prepared by adding 5 gm of crystal sugar to every 100 ml of sterile water (Abbott Laboratories, IL). The solution was sterilized using FDA approved techniques and osmolarity confirmed to be 292 mOsm using a 5500 vapor pressure osmometer (Wescor, Inc., Logan, UT). After completing the exposures, repeat blood and urine tests for electrolytes and osmolarity were carried out. Finally, subjects repeated the symptom questionnaire five days after the first visit, over the telephone. The pre-established criterion for discontinuing study participation was a decline in FEV1 by greater than 20% from baseline. Measurement of lung function ---------------------------- Spirometry was performed using a Vmax V6200 Autobox (Sensor Medics Corp., Yorba Linda, CA), according to guidelines published by the American Thoracic Society \[[@B34]\]. The spirometer was calibrated prior to each visit. Spirometry was performed on seated subjects who were using nose clips. Respiratory symptom score ------------------------- The amount of symptoms was assessed at baseline and after each exposure. Subjects scored chest tightness, shortness of breath, cough, headache, chills, muscle soreness, phlegm, nausea, stuffy nose, sneezing, and fatigue on a visual analog scale from 0--10 cm (0 being symptom-free and 10 being extreme amount) \[[@B31],[@B32]\]. Bronchoscopy and Bronchoalveolar lavage (BAL) --------------------------------------------- We also examined the effect of aerosolized xylitol on markers of inflammation in the airways. A separate group of subjects underwent bronchoscopy and bronchoalveolar lavage (BAL) according to American Thoracic Society standards at 30 minutes (n = 6), and 3 hours (n = 5) after exposure to 10 ml of aerosolized iso-osmolar xylitol \[[@B35]\]. BAL was performed by instilling two 20-ml aliquots of sterile normal saline into the lingula. The second aspirate was used for cytokine measurements. BAL fluid was filtered through two layers of sterile gauze to remove mucus and centrifuged for 10 minutes at 1500 rpm to separate cells. The cell pellet was washed twice in Hank\'s Balanced Salt Solution without Ca^++^and Mg^++^and suspended in complete medium, Roswell Park Memorial Institute (RPMI) tissue culture medium (Gibco/BRL, Gaithersberg, MD). Differential cell counts were determined with cytospin (Shandon, Pittsburgh, Pa) slide preparations by using Wright-Giemsa stain. The cell-free fluid was frozen at -70°C until required for cytokine assay. Cytokine measurements were performed using enzyme linked immunosorbent assays for IL-6 and LTC-4. IL-6 levels were determined by a Quantikine Human IL-6 ELISA kit (R&D Systems; Minneapolis, MN). The limit of detection of IL-6 is 0.70 pg/ml. LTC-4 (leukotriene) levels were determined by a leukotriene C4 EIA kit (Cayman Chemical; Ann Arbor, MI). The limit of detection of LTC4 is 10 pg/ml. LTC4 of BALs were extracted and concentrated with Cysteinyl-Leukotriene Affinity Sorbent (Cayman Chemical; Ann Arbor, MI). Statistical analysis -------------------- We studied ten subjects with a gradual increase in exposure dose in the pilot safety study. Differences were analyzed using t-test, Wilcoxon signed rank test, and one way and two-way repeated measures analysis of variance (ANOVA) as indicated. Ninety-five percent confidence intervals were calculated where appropriate. All analyses were performed using SAS version 8.2 (SAS Institute, NC) and at a 5% significance level. Results ======= Safety in mice -------------- Mice tolerated the exposures well without any visible distress. The corresponding volume of the 150-minute exposure was approximately 45 ml. Among naïve mice, exposure to xylitol resulted in no significant change in bronchial hyperresponsiveness compared to saline (Figure [1](#F1){ref-type="fig"}; n = 6/group; p = ns baseline and all concentrations of methacholine). A similar lack of difference between the saline- and xylitol-exposed mice was noted in their tidal volume and respiratory frequencies responses (data not shown). In a separate group of naïve mice that underwent Penh measurements periodically during exposure to saline or xylitol, no significant change was seen in Penh (Figure [2](#F2){ref-type="fig"}). We carried out similar studies on mice that had been sensitized to, and challenged with ovalbumin, a common murine model of asthma. No significant changes in methacholine responsiveness were observed (data not shown). Figure [3](#F3){ref-type="fig"} shows airway resistance measured invasively using the Flexivent system in naïve mice, OVA-sensitized/OVA-challenged mice after saline exposure and OVA-sensitized/OVA-challenged mice after xylitol exposure. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Effect of saline and xylitol exposure on methacholine responsiveness in naïve mice (n = 6/group). Panel A reflects methacholine responsiveness before and after saline exposure. Panel B reflects methacholine responsiveness before and after xylitol exposure. Error bars = SD. P-values of all comparisons are non-significant. ::: ![](1465-9921-5-13-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Effect of saline vs. xylitol exposure on Penh of naïve C57BL/6 mice (n = 6). The figure shows mean Penh values for mice exposed to saline (circles) and xylitol (squares). Errors bars = SD. p = 0.21. ::: ![](1465-9921-5-13-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Invasive airway resistance measurement in response to methacholine challenge in naïve and ova-challenged C57BL/6 mice (n = 2/group) using Flexivent system. The figure shows mean airway resistance for naïve mice (squares) ova-challenged mice (triangles). ::: ![](1465-9921-5-13-3) ::: Whole lung lavage showed no significant differences in lavage fluid cell count and differential due to xylitol exposure. Naïve mice exposed to saline or xylitol demonstrated, as expected, a macrophage-predominant response. In contrast, OVA-sensitized/-challenged mice were characterized by airway eosinophilia in both saline- and xylitol-exposed groups (Table [1](#T1){ref-type="table"}). In summary, aerosolized xylitol was well tolerated by naïve and hypersensitive mice with no significant effects on the airway physiology or composition of airway inflammatory cells. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Whole Lung Lavage Cell Count and Differential in Naïve and Ova-challenged Mice ::: Experimental Group Total Cell Count (×10^6^) Mean (SD) Differential Count (%) -------------------------------------------- ----------------------------------------- ---------------------------- ------ ------ ------ Naïve mice-saline 0.26 (0.8) 99.6 0.17 0.17 0.0 Naïve mice-xylitol 0.25 (0.7) 99.0 0.34 0.0 0.66 Ova-challenged mice -- saline exposed 0.96 (0.1) 20.0 3.6 14.0 62.2 Ova-challenged mice -- xylitol exposed 0.78 (0.08) 21.3 9.0 9.0 61.0 ::: Safety in human volunteers -------------------------- Table [2](#T2){ref-type="table"} shows the baseline characteristics of the ten subjects who underwent graded exposure to aerosolized xylitol as a part of the pilot study. Mean age was 29.1 yrs, and equal numbers of males and females were studied. None of the subjects dropped their FEV1 by ≥ 20%. The mean baseline FEV1 was 92% predicted (SD = 6.9% predicted). There was no significant change in FEV1 % predicted after any exposure in comparison with baseline (Figure [4](#F4){ref-type="fig"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Baseline Characteristics in Normal Volunteers ::: Subject No. Age Years Gender M/F Ethnicity Baseline FEV1 (% predicted) ----------------- --------------- ---------------- ------------------ --------------------------------- 1 41 F Caucasian 92 2 34 M Caucasian 85 3 48 M African American 87 4 22 M Caucasian 106 5 25 M Asian 95 6 20 F Asian 85 7 22 M Caucasian 91 8 20 F Caucasian 86 9 28 F Caucasian 100 10 31 F Caucasian 89 Mean 29 92 SD 9.5 6.9 ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Effect of exposure to nebulized saline and xylitol on spirometry in normal volunteers (n = 10). The figure shows mean FEV1 (% predicted) at baseline, after exposure to saline (10 ml), and xylitol (1, 5, and 10 ml). Errors bars = SD. p = 0.19. ::: ![](1465-9921-5-13-4) ::: As shown in Table [3](#T3){ref-type="table"}, xylitol exposure did not induce any significant change in electrolytes and osmolarity. No changes in vital signs or oxygen saturation were noted throughout the study. The most common symptom reported was stuffy nose after xylitol exposure, which occurred in five (50%) subjects after the 10 ml dose (Table [4](#T4){ref-type="table"}). The mean VAS score among the five subjects for stuffy nose was 3.5 cm. This symptom resolved within minutes after exposure was complete. Other less frequent side effects reported include, cough by two subjects (mean VAS score, 0.5), chest tightness by two subjects (mean VAS score, 1.0), and phlegm production by three subjects (mean VAS score, 1.5). All of these symptoms had resolved by day five of telephone follow-up. One subject noted hiccups half way through the final xylitol exposure, which resolved soon after the exposure was complete. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Laboratory Results pre and post Xylitol Exposure (n = 10) ::: Serum test Baseline Mean ± (SD) After 10 ml xylitol Mean ± (SD) p value ------------------------ -------------------------- ------------------------------------- ------------- Glucose, mg/dL 89 (3.8) 89 (9.1) 0.98 Osmolarity, mosm/k 292 (5.2) 292 (3.9) 0.98 Sodium, mEq/L 141 (1.4) 141 (2.6) 0.75 Bicarbonate, mEq/L 25 (1.2) 24 (1.9) 0.41 Anion gap, mEq/L 13 (1.2) 13 (1.2) 0.69 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Adverse Events Score (centimeters, mean ± SD) using Visual Analog Scale (1--10)\* ::: Symptom Baseline VAS score Change Post-saline Change Post-10 ml xylitol Change on day 5 follow-up ------------------------- ------------------------ ------------------------ ------------------------------- ------------------------------- Chest tightness 0 0 0.2 ± 0.4 0 Shortness of breath 0 0 0 0 Cough 0.25 ± 0.8 0.05 ± 0.15 0 0 Headache 0 0 0.2 ± 0.6 0 Chills 0 0 0 0 Muscle soreness 0.2 ± 0.6 0 0 -0.2 ± 0.6 Phlegm 0.2 ± 0.6 0 0.25 ± 0.4 0 Nausea 0 0 0 0 Stuffy/Runny Nose 0 0 0.65 ± 0.9† 0 Sneezing 0 0 0 0 Fatigue 0.1 ± 0.3 0 -0.1 ± 0.3 0 \P-values of all changes from baseline are \>0.05 except for stuffy nose after xylitol expsoure. † P-value = 0.03. ::: An additional 11 subjects underwent bronchoscopy and bronchoalveolar lavage following xylitol inhalation. The mean cell count in the BAL fluid at 20 minutes (n = 6) and 3 hours (n = 5) after xylitol exposure was 1.2 ± 0.07 million cells/ml and 2.94 ± 1.48 million cells/ml respectively. All cell preparations had between 95--100% alveolar macrophages. BAL IL-6 and LTC-4 levels after xylitol exposure were below 0.70 pg/ml and 10 pg/ml respectively at all time points. Discussion ========== Lower respiratory tract colonization is an important step in the pathogenesis of pulmonary manifestations of chronic diseases such as CF and dyskinetic cilia syndrome and certain acute clinical entities such as ventilator-associated pneumonia. There is a continuing need for simple, cost-effective, and safe intervention to decrease colonization of lower airways. Studies have shown that lowering the salt concentration of airway surface liquid can enhance innate immunity by increasing the potency of the natural antimicrobial peptides. In addition to increasing the activity of individual ASL factors, lowering the NaCl concentration also independently enhances synergistic interactions \[[@B36]\]. Thus, lowering the salt concentration could improve the antimicrobial activity of the ASL in two ways: increasing the individual action of the factors, and augmenting synergism between them. This double effect could amplify the impact of relatively modest reductions in salt concentrations. The mechanism of this low salt concentration augmentation of killing remains unclear. The most popular hypothesis is that in low salt concentrations, charged particles become less shielded, increasing the interaction between the cationic proteins and the negatively charged bacteria \[[@B6],[@B7],[@B37],[@B38]\]. Irrespective of the mechanism, this effect suggests a therapeutic strategy: lowering ASL salt concentrations should enhance bacterial killing. Xylitol, when applied to airways as an iso-osmolar agent, can potentially lower airway salt concentration and therefore lower bacterial colonization in chronic infections. In addition to having low transepithelial permeability, it has the added advantage of being poorly metabolized by bacteria. In recent years, many osmotic agents have been aerosolized into human airways for mucus clearance. However, there are reports of bronchospasm associated with their use. This is the first study to our knowledge to use xylitol in an aerosolized form. The main adverse effect reported from oral xylitol use was diarrhea when the dose exceeded 40--50 gm/day \[[@B39]\]. Intravenous xylitol has also been used as parenteral nutrition in the post-operative period for many decades. There have been no major changes in serum electrolytes with xylitol infusion \[[@B40]\]. Parenteral xylitol can cause minimal hyperuricemia, but without any pathophysiological consequences \[[@B41]\]. Though tolerated well in modest doses, large doses of xylitol administered intravenously have been reported to cause renocerebral oxalosis, with renal failure \[[@B42]-[@B45]\]. Before xylitol use in humans for prevention or reduction of airway colonization can be attempted, animal studies on safety as well as studies on healthy volunteers are required. We made calculations of the amount of xylitol to be delivered to the airway surface of an adult. Mercer, et al. \[[@B46]\] measured a total surface area from trachea to bronchioles of 2,471 cm^2^. The depth of ASL may vary from the trachea to the small bronchioles; if an average depth of 10 μm is estimated, the total ASL volume would be \~2.5 mL. Thus, if we assume a uniform aerosol distribution, administration of a total volume of 2.5 mL of 300 mM xylitol to the airways would be expected to lower the salt concentration in half simply by a dilutional effect. If the mean ASL depth were 20 μm, then 5 mL of delivered solution would be required. Because the solution is iso-osmotic, immediate, major osmotic shifts of water across the epithelium should not occur, which leads to dilution of the salt concentration. Moreover, with time, the volume and salt concentration may decrease due to Na^+^-dependent salt absorption, the osmotic effects of which are counterbalanced by xylitol in the ASL \[[@B13]\]. Our preliminary calculations for dosing for mice experiments were derived as follows; Mercer, et al. \[[@B46]\] also estimated the total airway surface area in rats, which was 27.2 cm \[[@B3]\]. Assuming an average depth of 10 μm, the total ASL volume would be \~27 μl. For a mouse, given an average weight of 25 gm, which is 1/12th of weight of a rat, the ASL volume is approximately 2.25 μ l. For a 50% dilution we have to deliver 2.25 μl of xylitol solution. Mice have an approximate 10% lung retention rate for the particle size we generated \[[@B47]\], which will require us to aerosolize 22.5 μl of xylitol. However, we do not have data on the airborne concentration of xylitol to which the mice were exposed. For the generation and exposure system employed, a reasonable approximation is that 5% of the solution nebulized into the mixing chamber was available for inhalation in the exposure chamber. Thus, we would need to deliver approximately 450 μl of xylitol solution to provide the desired 50% dilution of ASL. We exposed both normal and hypersensitive mice to a cumulative volume of 84 ml of iso-osmotic xylitol, which is at least a 2-log increase (187×) from the proposed dose. There was no significant change in airway resistance nor in bronchial hyperresponsiveness after xylitol exposure in naïve or hypersensitive mice. This study shows that aerosolization of iso-osmotic xylitol is likely to be safe and well tolerated by human volunteers. There was no change in spirometry, laboratory test results as well as BAL cytokine levels after xylitol exposure. Earlier studies have reported bronchial hyperresponsiveness with aerosolization of hypotonic and hypertonic solutions. Thus, aerosolization of iso-osmotic xylitol could be tested for prevention and treatment of airway colonization. There are several potential limitations with this study. The validity of body plethysmography as a measure of respiratory physiology in mice has been recently questioned \[[@B48],[@B49]\]. However, several studies have shown good correlation between airway inflammation and changes in Penh \[[@B50]-[@B52]\]. Since the human study is a true pilot study, we did not have preliminary data on adverse events for the aerosolized route to base our sample size calculation; given its relatively small size, we do not have the power to detect rare complications. Our human study was unblinded due to the sweet taste of xylitol, which all the subjects experienced. However, our main outcome, FEV1 is unlikely to be biased by knowledge of the exposure. Finally, this was a brief exposure study. Inhalational toxicology studies of safety of long-term exposure to animals looking at histopathology and laboratory data in addition to pulmonary function testing are required before clinical use can be instituted. Conclusions =========== In summary, our data indicate that iso-osmotic xylitol can be safely delivered by aerosol to normal volunteers. Studies of safety with long-term exposure to animals are required before human use can be attempted. This could lead to exciting interventions to enhance the innate immunity of airway epithelia. Abbreviations ============= ANOVA Analysis of Variance ASL Airway Surface Liquid CF Cystic Fibrosis FEV1 Forced Expiratory Volume in 1 second GSD Geometric Standard Deviation Penh Enhanced Pause VAS Visual Analog Scale BAL Bronchoalveolar Lavage Acknowledgements ================ We thank Dayna Depping and Tom Recker for assistance with laboratory procedures, the staff of the General Clinical Research Center (RR00059) for help with the human volunteer study, the volunteers, James Torner, PhD, Michael Welsh, Jamie Kesselring for assistance with manuscript preparation, the Animal Care Unit, the In Vitro*Cell Models Core \[supported by the National Heart, Lung and Blood Institute, the Cystic Fibrosis Foundation, and the National Institutes of Diabetes and Digestive and Kidney Diseases (DK54759)\], funded in part by the RDP (R458), and the SCOR grant from the NIH (HL61234-06), and the support of the Environmental Health Sciences Research Center (NIH/NIEHS P30 ES 05605).	PubMed Central	2024-06-05T03:55:48.336067	2004-9-16	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521689/", "journal": "Respir Res. 2004 Sep 16; 5(1):13", "authors": [ { "first": "Lakshmi", "last": "Durairaj" }, { "first": "Janice", "last": "Launspach" }, { "first": "Janet L", "last": "Watt" }, { "first": "Thomas R", "last": "Businga" }, { "first": "Joel N", "last": "Kline" }, { "first": "Peter S", "last": "Thorne" }, { "first": "Joseph", "last": "Zabner" } ] }
PMC521690	Background ========== Persistent asthma is an allergic disease characterised by airway inflammation (\[[@B1]-[@B5]\]) and hyperresponsiveness to external stimuli (\[[@B1]\]). Mouse models of allergic airway sensitisation are often used to elucidate the pathobiology of this disease (\[[@B6]-[@B8]\]). To date, a number of techniques have been used to measure changes in lung function in response to bronchoconstricting agents in murine models of allergic bronchopulmonary inflammation (see \[[@B6],[@B8],[@B9]\] for reviews). One method that has gained recent popularity is unrestrained barometric plethysmography, which uses a \'pseudo-flow\' measurement to derive a dimensionless parameter known as enhanced pause (Penh). There are now several publications in the literature which claim to have documented airway hyperresponsiveness in allergen-driven murine models based on methacholine induced changes in Penh. However, it has also been well documented that Penh does not correlate with changes in the physiology of the lung (\[[@B10]-[@B14]\]), especially in chronic disease states (\[[@B15]\]). In contrast, the low frequency forced oscillation technique (LFOT) is able to provide sensitive measurements of respiratory system input impedance (Zrs) in the mouse, that are partitioned into components representing airway and parenchymal compartments by fitting the constant-phase model (\[[@B16]-[@B18]\]). Using LFOT, Tomioka et al. (\[[@B17]\]) found that systemic sensitisation followed by three antigen challenges, one of the most common allergen models utilised in studies using Penh, resulted in hyperresponsiveness that was confined primarily to the tissue compartment of the lung. This has important implications for the interpretation of results obtained with Penh that have demonstrated mechanisms underlying allergic inflammation in mice given that a significant portion of the respiratory system hyperreactivity to MCh in human asthmatics is a result of the response of the conducting airways (\[[@B19]\]). One of the most common methods for inducing allergic bronchopulmonary inflammation in mice involves systemic sensitisation with a specific antigen and Th-2 skewing adjuvant, usually ovalbumin (OVA) adsorbed onto aluminium hydroxide (Alum), followed by airway challenge with the same antigen (\[[@B20]-[@B22]\]). However, considerable variations exist between studies in terms of the dose of antigen used during airway challenge. To date, a number of studies have found that airway hyperresponsiveness is increased by increasing the dose of antigen at challenge (\[[@B23]-[@B25]\]). However, these studies, which used different doses of antigen at challenge as part of a broader intervention protocol, have used Penh (\[[@B23],[@B24]\]) or a measure of total lung resistance (\[[@B25]\]) to examine the resulting changes in lung physiology. As yet, no studies have systematically examined the effect of the dose of antigen at challenge on the subsequent development of hyperresponsiveness using a technique that is able to partition the reactivity of the lungs into airways and tissue compartments. Hyperresponsiveness of the respiratory system to bronchoconstricting agents, and other outcome parameters such as those that reflect inflammation and allergic sensitisation, are often measured at different times post challenge. In an examination of the kinetics of hyperresponsiveness in an OVA model of allergic sensitisation in mice using a single dose of antigen at challenge, Tomkinson et al. (\[[@B26]\]) found that responsiveness to methacholine (MCh) is maximal 24 hours post challenge, has begun to resolve by 48 hours, and has returned to baseline levels beyond that time. The kinetics of responsiveness to MCh in other studies, however, are often overlooked and it is yet be determined if altering the dose of antigen at challenge has an influence on the timing of peak responsiveness to bronchoconstricting agents. The aim of this study was to systemically investigate the effect of antigen dose at challenge on the pattern of hyperresponsiveness to inhaled MCh in a murine model of allergic bronchopulmonary inflammation. Methods ======= Animals ------- 8 week old specific pathogen free female BALB/c mice were purchased from the Animal Resources Centre, Murdoch, Western Australia. Mice were housed in a controlled environment with a 12 hr light:dark cycle and provided with an OVA free diet and acidified water ad libitum. All experiments were approved by the Institutional Animal Ethics and Experimentation Committee. Sensitisation protocols ----------------------- Mice were sensitised by intraperitoneal (i.p.) injection with 20 μg of OVA (Sigma, St Louis, USA) suspended in 200 μL of Alum (Alu-gel-S, Serva, Heidelberg, Germany) on days 0 and 14. Mice were then challenged with either 1, 3 or 6 OVA (1% w/v in PBS) aerosols delivered with an ultrasonic nebuliser (UltraNeb^®^, DeVilbiss, Somerset, Pennsylvania) for 30 minutes on consecutive days starting at day 21 (Fig [1](#F1){ref-type="fig"}). Two additional groups of mice served as controls; a naïve group and a group sensitised with i.p. OVA and challenged with a single PBS aerosol using the protocol described above. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Timeline for sensitisation and data collection.Timeline for the protocols used to induce allergic bronchopulmonary inflammation and timing for bronchoalveolar lavage (BAL), serum IgE measurement and assessment of hyperresponsiveness to inhaled methacholine (MCh). Mice were systemically sensitised with two intraperitoneal injections of OVA/Alum on day 0 and 14, challenged with either 1 (A), 3 (B) or 6 (C) OVA aerosols (1%) for 30 minutes starting at day 21. ::: ![](1465-9921-5-15-1) ::: Respiratory mechanics --------------------- Changes in Zrs were measured using a modification of the low frequency forced-oscillation technique (LFOT) as described previously (\[[@B27]\]). Briefly, mice were anaesthetised with an i.p. injection of a solution containing xylazine (2 mg.mL^-1^, Troy Laboratories, NSW, Australia) and ketamine (40 mg.mL^-1^, Troy Laboratories, NSW, Australia) at a dose of 0.01 mL.g^-1^. Mice were tracheostomised with a 10 mm section of polyethylene tubing (1.27 mm OD: 0.86 mm ID) and ventilated (flexiVent, Scireq, Montreal, Canada) at 450 b.min^-1^with a tidal volume of 8 mL.kg^-1^and a positive end expiratory pressure (PEEP) of 2 cmH~2~O. The lung volume history of the mice was standardised prior to measurement of lung mechanics using two deep inflations and three P-V curves. The respiratory system input impedance (Zrs) was measured during periods of apnea using a 16 s signal containing 19 mutually prime sinusoidal frequencies ranging from 0.25 to 19.625 Hz. The constant phase model (\[[@B16]\]) was then fit to the real and imaginary parts of the Zrs spectrum allowing the calculation of airway resistance (R~aw~), tissue damping (G), tissue elastance (H) and hysteresivity (η) (\[[@B28]\]). Methacholine responsiveness --------------------------- Changes in respiratory mechanics following inhaled MCh were measured either 24 or 48 hrs after the last OVA aerosol. Following measurement of baseline Zrs, mice were exposed to a 90 s saline aerosol delivered with an ultrasonic nebuliser (UltraNeb^®^, Devilbiss, Somerset, Pennsylvania). Zrs was then measured every minute for the next 5 minutes. This aerosol procedure was repeated with 1/2 log~10~incremental doses of MCh from 0.1 to 30 mg.mL^-1^with Zrs measured every minute for at least 5 minutes after the aerosol until the parameters calculated from the constant phase model had peaked. Inflammatory cell counts ------------------------ Separate groups of mice, sensitised using the same protocol described above, were anaesthetised and tracheostomised 24 or 48 hrs after their last aerosol. BAL fluid was collected by slowly infusing and withdrawing a 1 mL aliquot of PBS containing BSA (bovine serum albumin, 20 mg.mL^-1^, CSL, Victoria, Australia) and lidocaine (35 mg.mL^-1^, Sigma, St Louis, USA) from the lungs three times. The BAL was then centrifuged at 2000 rpm for 4 mins. The supernatant was removed and the pellet resuspended in PBS. The cells were stained with trypan blue to determine viability and the total cell count (TCC) obtained by counting the cells with a haemocytometer. Differential counts were obtained from the cytospin sample, stained with Leishman\'s stain and examined using light microscopy. Three hundred cells were counted from each sample to determine the relative proportions of each cell type. Serum IgE --------- In a separate group of mice, serum samples were periodically collected for analysis of total IgE. An additional control group was included in the analysis of serum IgE consisting of mice sensitised with PBS/Alum. Sera were diluted 1:7.5 in Delfia Assay buffer (Wallac Oy, Turku, Finland). The diluted sera were analysed for the presence of total IgE by time-resolved fluorescence (TRF) assays. Briefly, 96-well plates (Nunc Maxisorp, Denmark) were coated overnight at 4° C with anti-mouse IgE (R35-72; BD PharMingen, San Diego, USA). Plates were blocked with 200 μl of 0.5% BSA in TRIS-HCl pH 7.4 for 1 hour at room temperature on a plate shaker. For all subsequent steps a volume of 50 μl per well was used and incubations were performed for 1 hour at room temperature unless otherwise indicated. Between steps, plates were washed five times with wash buffer (TRIS-HCl pH 7.8 Tween20). Mouse anti-TNP IgE (BD PharMingen, San Diego, USA) was used as an interassay standard. Biotinylated anti-mouse IgE (R35-118; BD PharMingen, San Diego, USA) was added to the wells at 2 μg.mL^-1^. Straptavidin-conjugated Europium (Wallac Oy, Turku, Finland) was incubated at 1:500 for 30 minutes and plates washed eight times thereafter. Delfia enhancement solution (Wallac Oy, Turku, Finland) was added and the plates were agitated on a shaker for 10 minutes prior to reading the fluorescence on a Wallac Victor 2 counter (Wallac Oy, Turku, Finland). The detection limit of this assay is approximately 100 ng.mL^-1^. Statistical analysis -------------------- Log~10~transformed inflammatory cell and immunoglobulin data were compared using ANOVA and Tukey\'s post-hoc test. Responses in R~aw~and G to inhaled MCh at the maximum dose used (30 mg.mL^-1^) were expressed as a percentage of the response to the saline aerosol and compared using non-parametric ANOVA on ranks and Dunn\'s post-hoc test. Responses in H were expressed as a percentage of the response to saline, log~10~transformed and compared using ANOVA and Tukey\'s post-hoc test. The threshold dose of MCh where there was a detectable change in R~aw~, G or H (termed sensitivity hereafter) was interpolated from the raw dose response curve as the upper limit of the 99% CI of the 5 measurements taken following the saline aerosol (Fig. [2](#F2){ref-type="fig"}). The sensitivity data were compared using ANOVA and Tukey\'s post-hoc test. All data were analysed using SigmaStat 2.03 and p values \< 0.05 were deemed to be significant. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Technique for sensitivity calculation.Schematic representation of the technique used for calculation of the threshold dose of MCh (sensitivity) required to induce a detectable increase in R~aw~, G and H. ::: ![](1465-9921-5-15-2) ::: Results ======= Methacholine responsiveness --------------------------- The degree and time of observed maximum MCh induced responses in R~aw~, G and H varied substantially between treatments (Fig. [3](#F3){ref-type="fig"}). A summary of statistical comparisons of sensitivity to MCh and percentage response to the maximum dose (30 mg.mL^-1^) between treatment groups and naïve mice is presented in Table [1](#T1){ref-type="table"}. Sensitisation followed by challenge with a single PBS aerosol did not cause an increase in sensitivity or maximum responsiveness to MCh compared to naïve mice. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Dose response curves to inhaled methacholine.Dose response curves (expressed as a % of the response to saline aerosol) for mice systemically sensitised with OVA/Alum and challenged via the airways with 1 (left), 3 (centre) or 6 (right) OVA aerosols. Mice were challenged 24 (●) or 48 (▲) hours after the last OVA aerosol. Dose response curves from naïve mice (○) are also shown. All data are expressed as mean ± SEM (n = 7--8). \* indicates significance (p \< 0.05 vs naïve mice; ANOVA on Ranks, Dunn\'s post-hoc for R~aw~and G; ANOVA, Tukey\'s post-hoc for H). ::: ![](1465-9921-5-15-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Summary of sensitivity and maximum responses to methacholine in airway and parenchymal lung compartments. Summary of the threshold dose (sensitivity) required to elicit a detectable increase in airway resistance (R~aw~), tissue damping (G) and tissue elastance (H) for naïve mice, mice systemically sensitised with OVA/Alum and challenged with PBS and mice systemically sensitised and challenged with OVA. Also shown is the percentage change in R~aw~, G and H in response to the maximum does of methacholine used (30 mg.mL^-1^). Data are presented as the mean (SEM). ::: Challenge Assessed after last aerosol (hr) Sensitivity - Threshold dose of MCh (mg.mL^-1^) Response at 30 mg.mL^-1^MCh ---------------- -------------------------------------- ----------------------------------------------------- --------------------------------- ------------ ------------ ------------ ---------- ------------- ------------ ------------- ------------- ------------- ------------- Raw G H Raw^§^ G^§^ H^§^ p\* p\* p\* p\* p\* p\* Naïve \- 0.54(0.14) \- 0.51(0.25) \- 0.10(0.02) \- 235.5(21.5) \- 138.8(6.1) \- 145.4(5.1) \- 1 PBS aerosol 24 and 48 pooled 0.55(0.30) ns 0.25(0.10) ns 0.06(0.01) ns 242.9(19.1) ns 141.5(4.7) ns 143.4(2.7) ns 1 OVA aerosol 24 0.09(0.03) 0.012 0.15(0.04) ns 0.05(0.01) ns 514.0(82.7) \<0.05 275.5(30.6) \<0.001 250.2(24.5) \<0.001 48 0.46(0.19) ns 0.18(0.05) ns 0.17(0.07) ns 326.4(32.6) ns 213.0(14.6) ns 179.3(12.6) ns 3 OVA aerosols 24 0.12(0.03) 0.012 0.35(0.20) ns 0.06(0.01) ns 271.2(43.1) ns 180.3(41.7) ns 178.9(39.9) ns 48 0.20(0.07) 0.034 0.16(0.04) ns 0.08(0.01) ns 348.8(46.0) ns 243.6(42.7) ns 233.0(31.6) 0.045 6 OVA aerosols 24 0.17(0.05) 0.019 0.12(0.05) ns 0.06(0.01) ns 456.7(43.4) \<0.05 358.2(88.4) \<0.05 298.5(52.2) 0.02 48 0.16(0.05) 0.034 0.15(0.06) ns 0.06(0.01) ns 396.0(23.3) \<0.05 304.8(28.5) \<0.05 295.6(30.6) 0.018 § expressed as a % of saline response \* vs naïve values ::: ### One OVA aerosol A single OVA aerosol was sufficient to induce a significant increase in MCh responsiveness in the airways, seen as both a lower threshold dose of MCh required to induce a response (increased sensitivity) and increased response at the 24 hour time point (Table [1](#T1){ref-type="table"}). In the parenchymal compartment, no increase in sensitivity was seen but a significant increase in maximal response was seen for both G and H. This heightened sensitivity and response had diminished, back to the level seen in naive mice, 48 hours after the OVA aerosol. ### Three OVA aerosols Three OVA aerosols resulted in significantly increased airway (but not parenchymal) sensitivity to MCh at both the 24 and 48 hour time points (Table [1](#T1){ref-type="table"}). However, there was no increase in maximum response at 24 hours in R~aw~, G or H and only an increased response in H after 48 hours but not R~aw~and G. ### Six OVA aerosols Six OVA aerosols resulted in both significantly increased airway sensitivity and maximal responses to MCh at 24 and 48 hours post-challenge. Increased maximal responses, but not increased sensitivity, were also seen in the parenchymal compartment at both the 24 and 48 hour time points. Inflammatory cell counts ------------------------ Challenge with a single PBS aerosol following systemic sensitisation with OVA did not cause a significant increase in TCC in the BAL (p = 0.552) compared to naïve mice (Fig. [4](#F4){ref-type="fig"}). There was, however, a significant increase in TCC in mice challenged with a single OVA aerosol (p = 0.032) and a further increase in TCC following 3 OVA challenges (p \< 0.001). Exposure to 6 OVA aerosols did not cause any further increase in TCC above levels observed in mice exposed to 3 OVA aerosols (p = 0.805) but remained significantly higher than mice challenged with 1 OVA aerosol (p \< 0.001). Time of sampling after the last aerosol with any of the protocols did not have a significant impact on TCC (p = 0.357). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Total cell counts from bronchoalveolar lavage.Total cell counts (TCC) from the bronchoalveolar lavage (BAL) of naïve BALB/c mice, mice systemically sensitised and challenge with OVA aerosols and mice systemically sensitised with OVA and challenged with PBS. Samples were collected 24 (grey) and 48 (black) hours after the last aerosol. Data are expressed as mean ± SEM (n = 5--6). Exposure to a PBS aerosol following antigen sensitisation did not cause an increase in TCC (p = 0.552). In contrast, a single OVA aerosol was sufficient to cause a significant increase in TCC (p = 0.032). Exposure to 3 OVA aerosols caused a further increase in TCC (p \< 0.001) but 6 OVA aerosols did not cause an increase in TCC beyond those observed in mice exposed to 3 OVA aerosols (p = 0.805). ::: ![](1465-9921-5-15-4) ::: The number of aeroallergen challenges also had a significant impact on the number of eosinophils (p \< 0.001) and macrophages (p \< 0.001) in the BAL. There were significant increases in the number of eosinophils in sensitised mice challenged with 1 (p = 0.032), 3 (p \< 0.001) and 6 (p \< 0.001) OVA aerosols (Fig. [5](#F5){ref-type="fig"}) compared to naïve mice. The numbers of eosinophils in the BAL of mice exposed to 3 and 6 aerosols were significantly higher than those exposed to a single OVA aerosol (p \< 0.001 and p \< 0.001 respectively) but were not significantly different from each other (p = 0.805). The number of macrophages in the BAL were also higher in mice exposed to 3 (p \< 0.001) and 6 (p \< 0.001) OVA aerosols compared to naïve mice. As with TCC, time of sampling after the last aerosol did not have a significant impact on the number of eosinophils (p = 0.357) or macrophages (p = 0.079) in the BAL. Low levels of neutrophils were observed in BALs from OVA challenged mice sampled at 24 hours but not in mice sampled 48 hours after the last OVA aerosol (Fig. [5](#F5){ref-type="fig"}). Lymphocyte numbers were not significantly elevated in the BALs from any of the treatment groups (data not shown). ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Differential cell counts from bronchoalveolar lavage.Differential cell counts from the bronchoalveolar lavage (BAL) of naïve BALB/c mice, mice systemically sensitised and challenge with 1,3 or 6 OVA aerosols and mice systemically sensitized with OVA and challenged with a single PBS aerosol. BALs were collected 24 and 48 hours after the last aerosol. Data are expressed as mean ± SEM (n = 5--6). There was a significant increase in the number of eosinophils (p = 0.032) in the BAL following a single OVA aerosol. Exposure to 3 or more OVA aerosols caused a further increase in the number eosinophils (p \< 0.001), compared to 1 OVA aerosol, and an increase in the number of macrophages (p \< 0.001) compared to naïve mice. There were neutrophils present in the BALs of some mice but only in those groups sensitised and challenged with OVA and only in BALs sampled 24 hours after the last aerosol. ::: ![](1465-9921-5-15-5) ::: Serum IgE --------- Total serum IgE was significantly increased at day 21 (p \< 0.001), 7 days after the second injection of OVA/Alum, compared to naïve mice (Fig. [6](#F6){ref-type="fig"}). In contrast, serum IgE levels at day 14, after a single injection, were not significantly elevated (p = 0.438) compared to naïve mice. The total serum IgE response to systemic sensitisation, in the absence of subsequent antigen aerosol challenge, peaked at day 22 and partially declined by day 27. However, this decrease was not statistically significant (p = 0.511). There was no further increase in the total serum IgE in mice that were sensitised and subsequently challenged with OVA aerosols compared to those that were only systemically sensitised (p = 0.842). Total serum IgE levels were not significantly greater in mice sensitised with PBS/Alum and challenged with OVA (data not shown). ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Total serum IgE obtained from time resolved fluorescence.Total IgE obtained from time resolved fluorescence assay of serum collected from systemically sensitised (i.p. OVA/Alum on day 0 and day 14) but not challenged with aerosolised antigen (white bars). The vertical bars represent total serum IgE from mice sensitised and challenged with either 1, 3 or 6 OVA aerosols. Serum samples from these mice were collected 24 (grey bars) and 48 (black bars) hours after the last aerosol. Data are expressed as mean ± SEM (n = 10). Two intraperitoneal injections of OVA/Alum were sufficient to induce increased levels total IgE by day 21 (p \< 0.001) compared to naïve mice. Exposure to OVA aerosol challenges did not cause a further increase in total IgE (p = 0.842). ::: ![](1465-9921-5-15-6) ::: Discussion ========== Varying the number of aeroallergen challenges in a systemically sensitised murine model of allergic bronchopulmonary inflammation altered the degree and timing of hyperresponsiveness to inhaled MCh. A single OVA challenge increased airway sensitivity to inhaled MCh 24 hours after the challenge, while sensitivity remained elevated for 48 hours after three and six challenges. OVA challenge did not increase parenchymal sensitivity at any level. In contrast to sensitivity measurements, the maximum response to 30 mg.mL^-1^MCh showed a variable pattern. A transient response was observed in both airway and parenchymal compartments after a single OVA aerosol. After 3 OVA aerosols significant increases were seen in the tissue compartment at 48 hours, while after 6 OVA aerosols an elevated response was seen in the airway and parenchymal compartments that persisted beyond 48 hours. There was a significant influx of inflammatory cells in the BAL in response to OVA aerosols, however, the presence of this inflammation did not always result in hyperessponsiveness to inhaled MCh. Murine models using 2 systemic allergen sensitisations followed by 3 aeroallergen challenges are prevalent in the literature (\[[@B20],[@B29]-[@B31]\]) and have been reported to demonstrate airway hyperresponsiveness to MCh. However, these studies have used enhanced pause (Penh), which is derived from unrestrained barometric plethysmography, to measure changes in lung physiology. As Penh cannot differentiate between constriction in the airways and changes in the tissue compartment of the lungs, it is impossible to tell where the responses to MCh are localised, if indeed they are true physiological responses (\[[@B10]-[@B14]\]). In contrast, our study, using 2 systemic sensitisations and 1,3 or 6 challenges, has demonstrated clear airway, tissue or mixed compartment responses to methacholine which is dependent on the number of aerosol challenges delivered. In our hands, the more common model of 2 systemic sensitisations followed by 3 OVA challenges resulted in increased responsiveness to the maximum dose of MCh that was confined to the tissue compartment of the lung. This finding is consistent with a previous study by Tomioka et al. (\[[@B17]\]), which also used a forced oscillation technique to measure changes in lung mechanics in OVA sensitised and challenged mice. The fact that the response was confined to the tissues is of interest as the aim of these models is to mimic the human asthmatic condition, in which a significant portion of reactivity of the lungs is localised in the conducting airways (\[[@B19]\]). This work emphasises the importance of measuring bronchoconstriction with physiological techniques capable of compartmentalising responses within the lungs. By varying the antigen dose at challenge we have revealed a system with the potential to allow investigation of transient or prolonged responsiveness to MCh that is localised in the airways, tissues, or both. Further investigation is needed in order to understand the mechanisms that are influencing the site of responsiveness. Typically, most human studies measure MCh responsiveness in terms of sensitivity as they report the concentration of MCh required to produce a 20% fall in FEV~1~. We have shown that it is possible to determine sensitivity to inhaled MCh in mice and that only the airway compartment shows heightened sensitivity following allergic sensitisation and challenge. While increased maximal responses can be seen in both airway and parenchymal compartments, depending on which model is used, no increase in parenchymal sensitivity is seen with any of the models we used. As such, these findings reinforce the value of using lung function techniques that are capable of assessing airway and parenchymal mechanics separately. Total serum IgE was significantly elevated following systemic sensitisation but was not increased by aerosol challenge. There was, however, a tendency for total serum IgE to decline by day 27 in mice that were systemically sensitised but not challenged with OVA aerosols, compared to mice additionally exposed to 6 OVA aerosols. It is possible that if the study had been extended to include further exposure to antigen over subsequent days, a difference would have been detected between mice that were only sensitised and mice that were sensitised and challenged. Given that antigen specific IgE and other immunoglobulin subtypes were not measured in this study, further work is required to characterise the effect of dose of antigen at challenge on the development of antibody responses to OVA in mice. The protocol used in the present study induced significant eosinophilia after a single airway challenge. The degree of eosinophilia increased with increasing number of airway challenges. This finding is consistent with several previous studies using similar protocols to induce allergic inflammation in the lungs of mice (\[[@B20],[@B29]-[@B31]\]). While the level of activation of the eosinophils was not measured in the present study, the 61% eosinophilia found after 6 OVA aerosols was much higher than those that are typically found in human asthmatics (\[[@B32]\]). Given the significant and prolonged parenchymal response to inhaled methacholine following 6 OVA aerosols and the level of eosinophilia present, it is likely that this model more closely parallels an allergic alveolitis (\[[@B33]\]) than the airway inflammation commonly seen in humans. In recent studies there has been some focus on the association, or lack thereof, between indicators of systemic sensitisation, such as the levels of serum antibodies, airway inflammation and AHR (\[[@B34]\]). In a review of the role of IgE in the induction of eosinophilic airway inflammation and AHR, Hamelmann et al. (\[[@B35]\]) concluded that systemic methods of sensitisation resulted in high levels of IgE and eosinophilic airway inflammation in BALB/c mice. In these models, AHR was determined to be dependent on eosinophils but not IgE. However, the results of our study, which uses a similar protocol to those reviewed by Hamelmann et al. (\[[@B35]\]), show that the presence of eosinophils did not always coincide with an increase in responsiveness to MCh. Three OVA aerosols resulted in a significant eosinophilia after 24 hours but an increase in the response to the maximum dose of MCh was not evident until 48 hours post challenge. In contrast, a single OVA challenge resulted in hyperresponsiveness to MCh that had resolved by 48 hours while the levels of eosinophils remained significantly elevated. The levels of total serum IgE were equivalent across all challenge doses suggesting that, while the presence of IgE may be necessary to initiate the allergic response, its presence at a particular measurement time point does not necessarily relate to the presence of hyperresponsivenss. Conclusions =========== The findings of the present study demonstrate the significant impact of changing antigen challenge dose in a murine model of allergic bronchopulmonary inflammation. Given the variability of the inflammatory profile and characteristic responses observed in this study, it is clear that investigators must carefully characterise their allergen-driven murine models to ensure the model used contains the characteristic of interest. Future studies need to be directed at understanding the mechanisms that underlie airway and parenchymal hyperresponsiveness post antigen challenge. Authors\' contributions ======================= GRZ carried out the animal studies and drafted the manuscript. CvG carried out the IgE analysis and assisted in the interpretation of results and editing the manuscript. PAS assisted in the interpretation of results and editing the manuscript. PGH assisted in the conceptualisation of the study and interpretation of the results. PDS and DJT assisted in the conceptualisation of the study, interpretation of the results and editing the manuscript. Acknowledgements ================ This project was supported by a National Health and Medical Research Council Program Grant \#211912. We also thank Sam Gard for preparation of the BAL samples and total cell counts.	PubMed Central	2024-06-05T03:55:48.338871	2004-9-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521690/", "journal": "Respir Res. 2004 Sep 23; 5(1):15", "authors": [ { "first": "Graeme R", "last": "Zosky" }, { "first": "Christophe", "last": "von Garnier" }, { "first": "Philip A", "last": "Stumbles" }, { "first": "Patrick G", "last": "Holt" }, { "first": "Peter D", "last": "Sly" }, { "first": "Debra J", "last": "Turner" } ] }
PMC521691	Background ========== Oxidative stress leading to the overproduction of free radicals in the lungs is present in many clinical situations. Such clinical settings include acute respiratory distress syndrome (ARDS), infants of prematurity going on to develop bronchopulmonary dysplasia (BPD), pathogenesis of chronic obstructive pulmonary disease (COPD), asthma, cystic fibrosis, ischemia-reperfusion injury, drug-induced lung toxicity, cancer and aging \[[@B1]-[@B4]\]. Although the use of oxygen may be clinically indicated in hypoxemic situations, one must consider the potential long-term toxic side effects. For example, we know that oxygen creates cellular damage by a variety of mechanisms. Normal cellular metabolism of oxygen involves the transfer of electrons from NADH to O~2~molecules to form water (H~2~O). At normal partial pressure, 95% of oxygen molecules (O~2~) are reduced to H~2~O and 5% are partially reduced to toxic byproducts by normal metabolism in the mitochondria \[[@B5]\]. These metabolites include the superoxide anion (O~2~^-^), hydrogen peroxide (H~2~O~2~), and hydroxyl radicals (^•^OH) all of which make up what are known as Reactive Oxygen Species (ROS) \[[@B6]\]. Exposure to conditions of hyperoxia as well as ionizing radiation (IR) leads to increased amounts of these ROS and their damaging effects. ROS are known to attack the lipids, proteins, and nucleic acids of cells and tissues \[[@B5],[@B7]\]. Lipids, including pulmonary surfactant, react with ROS to produce lipid peroxides, which cause increased membrane permeability, inactivation of surfactant, and inhibition of normal cellular enzyme processes. Proteins reacting with ROS result in decreased protein synthesis due to inhibition of ribosomal translation or destruction of formed proteins. This ultimately leads to inactivation of intracellular enzymes and transport proteins resulting in impaired cellular metabolism and accumulation of cellular waste products. Lastly, ROS cause damage to nucleic acids by leading to modified purine and pyrimidine bases, apurinic (AP) /apyrimidinic sites, and DNA protein cross-links which can lead to single strand breaks \[[@B8]\]. Several defense mechanisms exist to combat the damaging effects of ROS. Intracellular enzymatic systems include superoxide dismutase which eliminates the superoxide anion, catalase which catalyzes the reduction of H~2~O~2~directly to H~2~O without the production of the hydroxyl radical, and glutathione peroxidase which directly reduces H~2~O~2~and lipid peroxides. Free radical scavengers, which stop free radical chain reactions by accepting electrons, include α-tocopheral (vitamin E), ascorbic acid (vitamin C), niacin (vitamin B), riboflavin (vitamin B~2~), vitamin A, and ceruloplasmin \[[@B1],[@B2],[@B9]\]. These systems usually provide enough protection against oxygen metabolism under normal conditions, but may become depleted under conditions of increased oxidative stress \[[@B7],[@B10]\]. The defense mechanism of interest in this paper involves the repair of oxidative damage through the human DNA base excision repair pathway (BER). BER is the most important cellular protection mechanism that removes oxidative DNA damage \[[@B11]\]. Damaged bases are excised and replaced in a multi-step process. Lesion-specific DNA glycosylase repair genes initiate this process. After removal of the damaged base, the resulting AP site is cleaved by AP-endonuclease generating a 3\'OH and 5\'deoxyribose phosphate (dRP). β-polymerase, which possesses dRPase activity, cleaves the dRP residue generating a nucleotide gap and then fills in this single nucleotide gap. The final nick is sealed by DNA ligase \[[@B12]-[@B14]\] (Figure [1A](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Base excision repair pathways for Oxidative DNA damage.(A) BER pathway demonstrating repair of 8-oxoG by the repair enzymes hOgg1and hNTH. (B) hOgg1, hMYH, and hMTHand their respective repair function. ::: ![](1465-9921-5-16-1) ::: The oxidative repair genes that we have analyzed in this study include 8-oxoguanine DNA glycosylase (hOgg1), human Mut Y homologue (hMYH), human Mut T homologue (hMTH), and endonuclease III (hNTH) all of which are present in human cells and involved in the protection of DNA from oxidative damage. The repair enzyme hOgg1is a purine oxidation glycosylase that recognizes and excise 8-oxoguanine lesions (GO) paired with cytosine. GO can pair with both cytosine and adenine during DNA replication \[[@B15]\]. If repair of C/GO does not occur, then G:C to T:A transversions may result \[[@B5],[@B15]-[@B17]\]. The repair enzyme hMYHis an 8-oxoguanine mismatch glycosylase that removes adenines misincorporated opposite 8-oxoG lesions that arise through DNA replication errors \[[@B5],[@B18]-[@B20]\]. The repair enzyme hMTHhydrolyzes oxidized purine nucleoside triphosphates such as 8-oxo-dGTP, 8-oxo-GTP, 8-oxo-dATP, and 2-hydroxy-dATP, effectively removing them from the nucleotide pool and preventing their incorporation into DNA (Figure [1B](#F1){ref-type="fig"}) \[[@B21]\]. Lastly, the repair gene endonuclease III (hNTH) is a pyrimidine oxidation and hydration glycosylase that recognizes a wide range of damaged pyrimidines \[[@B22]\]. hNTHhas also been shown to have a similar DNA glycosylase/AP lyase activity that can remove 8-oxoG from 8-oxoG/G, 8-oxoG/A, and 8-oxoG/C mispairs \[[@B23],[@B24]\]. Subsequent steps following hNTHare identical to those following hOgg1(Figure [1A](#F1){ref-type="fig"}). A previous study has shown that over-expression of the DNA repair gene hOgg1leads to reduced hyperoxia-induced DNA damage in human alveolar epithelial cells \[[@B25]\]. The primary goal of our present study was to compare the protective effects of the four main lesion-specific DNA glycosylase repair genes by individually over-expressing each in lung cells and determining which of these provides the greatest degree of protection under conditions of increased oxidative stress. Methods ======= Cell Culture ------------ The human alveolar epithelial cell line A549 (58 year old Caucasian male), was purchased from ATCC Cat No CCL-185. The cells were grown in DMEM (Gibco, Grand Island, NY) supplemented with 10% fetal bovine serum (FBS) (HyClone, Logan, UT) and penicillin (100 U/ml)/streptomycin (100 μg/ml) (Gibco, Grand Island, NY). Passaging of cells was performed every 3--4 days with cells grown to 80% confluency in a 10 cm cell culture dish (Corning Incorporated, Corning, NY). Cells were kept at 37°C in a humidified, 5% CO2 incubator. Retroviral Vector Construction ------------------------------ The retroviral vector pSF91.1, a gift from Dr. C. Baum from the University of Hamburg in Germany, was constructed with an internal ribosome entry site (IRES) upstream to the gene expressing enhanced green fluorescent protein (EGFP) as previously described \[[@B26]\]. Four DNA repair genes were individually ligated into the retroviral vector pSF91.1. ### hOgg1-6pcDNA3.1 was initially amplified by PCR by primers to introduce a kozak sequence at the 5\' end \[[@B27]\]. Digestion of this product with EcoRI and SalI was performed and then hOgg1was subcloned into digested plasmid vector pSF91.1, with T4 DNA ligase. DNA sequencing was performed to confirm integrity of the hOgg1gene. ### hMYH/PGEX4T-1 and hMTH/PGEX4T-1 hMYHwas a gift from Dr. A. McCullough (University of Texas Medical School, Galveston, TX) and hMTHwas cloned in Dr. Kelley\'s lab. Plasmid DNA was prepared as above by digestion with EcoRI and SalI and ligated into pSF91.1 as above and sequencing was performed to confirm integrity of the genes. ### PGEX-6PI-hNTH1-wild type this gene was a gift from Dr. S. Mitra (University of Texas Medical School, Galveston, TX). Digestion with BamHI and SalI was performed and the hNTH1-wtfragment was ligated into the empty plasmid vector PUC18. The hNTH1-wtfragment was then excised with both sides flanked by EcoRI restriction sites and ligated into pSF91.1. Proper orientation of the gene was confirmed and sequencing was performed to determine the integrity of the gene. Retroviral Production and Infection ----------------------------------- DH5α competent cells (Life Technologies, Gaithersburg, MD) with each of the five DNA repair genes were grown in LB-broth with ampicillin (Sigma, St. Louis, MO). Plasmid DNA was prepared and used to transfect phoenix amphotropic cells, from the Nolan Lab (Stanford University Medical Center, San Francisco, CA), grown to \~80% confluency. On the second day sodium butyrate was added to each plate and incubated at 37°C for 6 hours. Fresh DMEM supplemented with FBS and penicillin/streptomycin was added and the plates were incubated at 33°C. Viral supernatant was collected 24 and 48 hrs later, filtered through a 0.45 μm acrodisc syringe filter (Pall Corporation, Ann Arbor, MI) and frozen at -80°C for later use. Retroviral titers were determined by fluorescent-activated cell sorter (FACS) analysis. Titers of viral supernatant were 8 × 10^5^to 1.2 × 10^6^particles/ml \[[@B26]\]. 2.5 × 10^5^A549 cells were suspended with the viral supernatant and plated in 1 well of a 6-well plate along with polybrene (Sigma, St. Louis, MO). This exposure was performed 6 hours per day for three days. At approximately five days from the beginning of the infection, the infected cells were analyzed using flow cytometry and sorted for EGFP expression. Western Analysis ---------------- Cell pellets of sorted cells were resuspended in NuPage buffer (Invitrogen, Carlsbad, CA) and protein concentrations were determined using the DC protein assay (Bio-Rad, Hercules, CA). 20 ug of protein were loaded into individual lanes of a NuPage Bis-Tris Gel (Invitrogen, Carlsbad, CA). The gel was then transferred to nitrocellulose paper (Osmonics Inc, Gloucester, MA). The membranes were then blocked with 1% blocking solution (Roche Diagnostics, Indianapolis, IN) for 1 hour at room temperature and then incubated overnight at 4°C with rabbit polyclonal antibodies to hOgg1(Novus Biologicals, Littleton, CO), hMTH(Novus Biologicals, Littleton, CO), hMYH(Oncogene Research Products, Darmstadt, Germany) and hNTH(Proteintech Group Inc, Chicago, IL) all at a dilution of 1:1000 except hNTHwhich was diluted 1:2500. They were then washed 2 times with TBST and 2 times with 0.5% blocking solution, 10 minutes per wash. The membranes were incubated with anti-rabbit secondary antibodies at 1:1000 for 1 hour at room temperature. Lastly, the membranes were washed 4 times with TBST, 15 minutes per wash. The membranes were briefly soaked in BM chemiluminescence blotting substrate (Roche Diagnostics, Indianapolis, IN) and then exposed to high performance autoradiography film (Amersham Biosciences, Buckinghamshire, England). Kodak Digital Science 1D Image Analysis software was utilized to quantify the region of interest (ROI) band mass of individual bands on films where visualized differences were detected. Hyperoxic Exposure ------------------ Sorted EGFP positive A549 cells infected with the above DNA repair genes were counted and seeded into 96-well plates at a density of 1000 cells/well, 6 wells per gene. Six hours after seeding, individual plates were placed into an oxygen chamber supplied by Dr. L. Haneline (Wells Center for Research, Indianapolis, IN) located in a 37°C incubator. The oxygen chamber was then infused with 95% O~2~and 5% CO~2~. Individual plates were removed after 12, 24, 48, and 72 hours of exposure. Control A549 cells were incubated in a normal 37°C humidified-5% CO~2~incubator. O~2~concentrations were monitored with a MAXO~2~analyzer (Maxtec, Salt Lake City, UT). Four days from the beginning of the exposure, cells were assessed for cell growth/survival using the sulforhodamine B assay (SRB assay). Sulforhodamine B Assay ---------------------- The SRB assay (Sigma, St. Louis, MO), developed by the National Cancer Institute, provides a sensitive measure of drug-induced cytotoxicity through a colorimetric endpoint that is non-destructive, indefinitely stable, and visible to the naked eye. This assay was used to assess the cell growth/survival of over-expressed cells \[[@B28]\]. Cold 10% TCA was used to fix the cells to the plate. After incubation for 1 hour at 4°C, the individual wells were rinsed with water. After air-drying, SRB solution was added to each well and cells were allowed to stain for 20--30 minutes. 1% acetic acid wash was used to rinse off unincorporated dye. Incorporated dye was then solubilized in 100 μl per well of 10 mM Tris. Absorbance was measured by a tunable microplate reader (Molecular Devices, Sunnyvale, CA) at a wavelength of 565 nm. Background absorbance measured at 690 nm was subtracted from the measurements at 565 nm. Irradiation and H~2~O~2~Exposure -------------------------------- Sorted EGFP positive A549 cells were seeded into 96-well plates at a density of 1000 cells/well. Six hours after seeding, individual plates were then exposed to radiation at doses of 250, 500, 1000, and 1500 Rads or 0.2 mM, 0.4 mM, and 0.6 mM H~2~O~2~(Sigma, St. Louis, MO). All plates including control plates were then placed into a 37°C humidified-5% CO~2~incubator. Four days after exposure, cells were fixed and assessed for cell growth/survival by the SRB assay. Natural Cell growth ------------------- Sorted EGFP positive A549 cells and wild type cells were seeded individually onto four 96-well plates at 1000 cells/well. All the plates were placed into a 37°C humidified-5% CO~2~incubator. Every 24 hours for 4 days, 1 plate was removed and the cells were fixed and analyzed by the SRB assay looking at cell growth under non-toxic conditions. Growth curves and exponential growth equations were determined to look at the doubling time (DT) of cells infected with each repair gene of interest compared to vector infected and uninfected wild type cells. Statistics ---------- All drug exposure experiments were performed at least three times and individual drug doses included 6--8 wells for each group of infected cells. Analysis of cell growth and exponential growth equations were determined using Microsoft Excel. All experiments involving drug exposures were normalized to the zero dose. Data are expressed as means ± SE. The significance of differences were calculated using the paired Student\'s ttest with significance being accepted for p \< 0.05. Results ======= Retroviral Constructs --------------------- The DNA repair genes hOgg1, hMYH, hMTH, and hNTHwere ligated into the retroviral vector pSF91.1 (figure [2](#F2){ref-type="fig"}). This vector, derived from a murine stem cell virus backbone, along with each individual repair gene, was used for transfection of phoenix amphotropic cells. Viral supernatant was then collected and used to stably infect A549 cells. A heterogeneous population of A549 cells expressing EGFP was sorted so all cells used for experiments contained the genes of interest integrated into their DNA (data not shown). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Retroviral vector pSF91.1.Depiction of the retroviral vector utilized in these experiments demonstrating restriction sites and location of entry of the gene of interest between the LTR and the IRES. ::: ![](1465-9921-5-16-2) ::: Repair Gene Expression ---------------------- Western blot analysis was performed on sorted cells in order to verify over-expression of the four genes of interest. hOgg1, hMYH, hMTH, and hNTHwere all detected at their correct position on western blots (data not shown). Western analysis was also utilized to assess whether over-expression of each individual repair gene resulted in altered endogenous expression of the other repair genes under both non-toxic and toxic conditions (24 hrs of 95% O~2~and 1000 Rad). Cells over-expressing the repair genes hOgg1, hMYH, hMTH, and hNTHdid not lead to altered expression of the other endogenous repair genes under the above conditions when compared to each other or pSF91.1 vector control cells (Figure [3A,3B,3C](#F3){ref-type="fig"} and [3D](#F3){ref-type="fig"}). hOgg1\'s endogenous expression was below the level of detection. The pattern of endogenous expression of hNTHwas consistent for each condition when comparing cells over-expressing hOgg1, hMYH, hMTH, and pSF91.1. Reduced expression of hNTHafter exposure to 95% O~2~was noted. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Western analysis of A549 cells over-expressing individual repair genes and effect on endogenous glycosylase level.(A) Endogenous expression of hOgg1 was not altered in A549 cells over-expressing any of the other repair genes when analyzed after non-toxic and toxic exposures. hOgg1protein was not detectable for any of the cells under the above conditions when compared to cells over expressing hOgg1. (B) and (C) Endogenous expression of hMTH and hMYH respectively also were not altered in A549 cells over-expressing any of the other repair genes when analyzed after non-toxic and toxic exposures. (D) Endogenous expression of hNTH was analyzed under non-toxic and toxic conditions in A549 cells over-expressing the other repair genes. Reduced expression of hNTH was observed equally with all of the other genes after exposure to 95% O~2~. Endogenous expression of all four genes was equivalent under the above conditions in vector control cells; pSF91.1 (data not shown). ::: ![](1465-9921-5-16-3) ::: Lastly, we assessed endogenous expression of each individual repair gene in cells infected with pSF91.1 following non-toxic and toxic conditions (24 hrs of 95% O~2~and 1000 Rad) at 24 and 48 hrs after the onset of exposure. Endogenous hMYHand hMTHwere expressed to the same degree. hOgg1\'s endogenous expression was below the level of detection using western analysis (results not shown). When analyzing endogenous hNTHexpression, it was noted that hyperoxia at 24 hrs and 48 hrs resulted in reduced protein expression by 93% and 64% respectively. There also was a small increase in expression of hNTHnoted after 1000 Rad one day post exposure that was back to baseline by two days post exposure. ROI band mass quantification demonstrated this finding (Figure [4A](#F4){ref-type="fig"} and [4B](#F4){ref-type="fig"}). Two or more replicates were performed for each western analysis to determine consistency of the results. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### *Western analysis of endogenous hNTHrepair gene after exposure to O~2~and IR.(A) Analysis of hNTHexpression in A549 vector control cells following O~2~or IR treatment. The ROI band mass mean intensity was calculated for individual bands and hNTHexpression was normalized to the corresponding actin loading control. (B) Graph of ROI band mass normalized to the pSF91.1 zero dose. ::: ![](1465-9921-5-16-4) ::: Protection from Hyperoxia and Radiation --------------------------------------- A549 cells expressing hMYHdemonstrated increased survival after exposure to conditions with elevated levels of oxygen compared to cells expressing only the pSF91.1 vector (Figure [5A](#F5){ref-type="fig"}). Results were highly significant at all time points except after 12 hours O~2~where it almost reached a highly significant value. The differences between pSF91.1 and hMYHvaried from 12% after 12 hours O~2~exposure to 7% after 72 hours O~2~exposure. A549 cells expressing hMYHalso demonstrated increased survival after exposure to all doses of radiation in comparison to pSF91.1 (Figure [5B](#F5){ref-type="fig"}). These results were also highly significant at all doses of radiation except at 250 Rads where it almost reached a highly significant value. The differences between pSF91.1 and hMYHvaried from 12%--14% for all doses of radiation. Also noted in these experiments was that vector control cells demonstrated no significant difference in survival at all doses of O~2~and radiation in comparison to wild type A549 cells. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Cell survival analysis following O~2~, IR, and H~2~O~2~treatments.A549 cells over-expressing hOgg1, hMYH, hMTHand hNTHfollowing (A) O~2~, (B) IR, and (C) H~2~O~2~. Brackets indicate statistical significance at \ p \< 0.05 and \\ p \< 0.001 compared to pSF91.1 at each individual dose for 1 representative experiment. ::: ![](1465-9921-5-16-5) ::: Experiments looking at the effects of H~2~O~2~on cells expressing the repair genes did not demonstrate increased survival for any of these repair genes when compared to vector control cells (Figure [5C](#F5){ref-type="fig"}). This data demonstrates that over-expression of hMYHhas the ability to improve cellular survival under conditions of hyperoxia and radiation but may not be able to overcome the toxicity of H~2~O~2~. Cell Growth ----------- Cell growth under normal conditions was ascertained to determine if over-expression of any of the repair genes caused an alteration in the growth of cells in the absence of oxidative stress. Wild type A549 cells and cells expressing pSF91.1, hNTH, hOgg1, and hMTHappeared to grow at similar rates with doubling times within the same range. A549 cells expressing hMYHdid show a slower growth rate that resulted in significant differences in cell number by day 3. The calculated doubling time for the cells over expressing hMYHis \> 3 hrs longer than the cells with the other repair genes and vector alone (Figure [6](#F6){ref-type="fig"}). This slowing of growth may allow for more time to repair DNA damage, ultimately leading to increased cell survival following oxidative stress. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Cell growth curve and associated doubling times (DT).A549 cells over-expressing hMYHgrow at a slower rate in comparison to all other cells under non-toxic conditions resulting in a prolongation of the doubling time. Of note, all other over-expressed cells have approximately the same doubling time as wild type A549 cells. Statistical significance noted at \\ p \< 0.001 compared to pSF91.1 for 1 representative experiment. ::: ![](1465-9921-5-16-6) ::: Discussion ========== Oxidative stress to the lung leads to cellular DNA damage as evidenced by the release of specific gene products known to regulate DNA base excision repair pathways such as p53 and p21 \[[@B29]-[@B31]\]. Alterations in pro-inflammatory mediators, transcription factors, and other related gene products are also observed \[[@B32]\]. This injury has been shown to be associated with features of both cellular necrosis and apoptosis \[[@B33]-[@B35]\]. The resultant cellular inflammation and death from oxidative stress has a dramatic impact on the outcome of patients in the clinical setting \[[@B7],[@B36]\]. Most of our current clinical therapy towards oxidative stress in the lung involves both supportive measures and prevention. Research dealing with oxidative lung injury has focused mainly on enhancing antioxidant enzymatic processes and free radical scavengers \[[@B37]-[@B40]\]. The ability to alter cellular survival by increasing specific DNA repair mechanisms may add another approach to the treatment of oxidant-mediated lung injury. Many investigators have used hydrogen peroxide as a substitute for hyperoxia since it is known to be one of the metabolites produced by the metabolism of oxygen. ROS such as H~2~O~2~and those produced by hyperoxia clearly lead to DNA damage but questions exist as to whether H~2~O~2~leads to the same deleterious effects upon DNA as hyperoxia. Analysis of our growth curves after exposure to H~2~O~2~in comparison to hyperoxia and IR clearly indicates that cellular protection by oxidative DNA repair genes is specific to the agent used. Because no protection was observed with over-expression of any of the repair genes following exposure to H~2~O~2~, we speculate that the damage it causes is dissimilar. It may be that its damage not only involves oxidized bases, but may also include other forms of DNA, lipid, and protein damage that are not corrected by oxidative DNA repair genes. Alternatively, the amount and type of damage evoked by H~2~O~2~could be beyond that which can be corrected by over-expressing these repair genes. Another form of stress known to induce damage through the formation of ROS is IR. Radiation induced free radical damage to DNA has substantial overlap with that of oxidative damage \[[@B41]-[@B43]\]. The protection provided by specific oxidative DNA repair genes under conditions of IR, was notable throughout our experiments only with the repair enzyme hMYH. The primary agent utilized to induce the formation of ROS was an oxygen rich environment. The use of oxygen as a stressor leading to the formation of ROS, offers a distinct advantage over IR and H~2~O~2~by mimicking the clinical situation where constant exposure to hyperoxia leads to cumulative cellular damage which further compromises repair. We determined that survival of A549 cells was also enhanced to a small degree with increased expression of the repair enzyme hMYH. This was an unexpected finding as we anticipated the repair gene hOgg1would demonstrate the greatest protection in response to oxidative stress based on previous studies, however these experiments utilized the colony forming assay (CFA) to detect improvements in survival \[[@B25]\]. Additionally, the CFA may provide different results compared to the SRB assay, which allows for growth analysis over a shorter window of time. Furthermore, their study did not look at the repair enzyme hMYHand its impact on survival. Another study has investigated the repair function of hMYHin MYH-deficient murine cells. It was demonstrated that transfection of the MYH-deficient cells with a wild-type MYH expression vector increased the efficiency of A:GO repair \[[@B44]\]. An interesting observation noted while doing our experiments lead us to look at individual growth characteristics of cells over-expressing each of the oxidative repair enzymes. Cells over-expressing the repair enzyme hMYHclearly grow at a slower rate when compared with the other enzymes. The mechanism behind this is not understood at this point in time. The repair action of hMYHis known to remove adenines misincorporated opposite 8-oxoG lesions. This lesion occurs when a C/GO lesion is allowed to replicate before being corrected by hOgg1. Repair by hMYHis not a final corrective measure. The product of hMYHactivity is the lesion C/GO, which allows hOgg1to have another opportunity to remove 8-oxoG opposite cytosine. We know that A549 cells possess the hOgg1gene based on a previous study demonstrating the presence of this gene after amplification by genomic PCR \[[@B45]\]. We also have demonstrated endogenous activity of hOgg1in A549 cells by using an 8-oxoguanine bioactivity assay. Therefore, our explanation of these results is that the slowed growth created by hMYHmay provide a wider window of opportunity for the repair process to take place, which ultimately grants endogenous hOgg1another opportunity to remove the 8-oxoG lesion created by oxidative stress. As noted in the methods section, the SRB assay provides a sensitive measure of drug-induced cytotoxicity that is used to assess cell proliferation/survival. The reduced cell proliferation of A549 cells over-expressing hMYH under non-toxic conditions may likely underestimate the magnitude of the protective effect of this particular repair enzyme. This may in fact make the results even more significant. Recent studies have discovered hereditary variations of the glycosylase hMYHthat may predispose to familial colorectal cancer \[[@B46],[@B47]\]. Others have looked for hMYHvariants in lung cancer patients and have not identified any clear pathogenic biallelic hMYHmutations or an over-representation of hMYHpolymorphisms \[[@B47]\]. The A549 cell line has not demonstrated somatic mutations in hMYH, but a single nucleotide polymorphism (SNPs) has been noted \[[@B45]\]. The impact on function by this SNP is unknown. It would appear that the function of hMYHis very important in preventing somatic mutations leading to cancer in the gastrointestinal tract. Although studies to date have not demonstrated this same relationship with lung cancer, we do know that the lungs are subjected to large quantities of ROS under certain conditions as discussed earlier. The formation of mutations from oxidative stress does have other deleterious effects on cells including cellular death by necrosis and apoptosis. Tissue viability is dependent upon mutation correction and replication of the surviving cells to replace those that have died. The ability to enhance cellular survival, after specific oxidative exposures, is evident after increased production of the hMYHrepair gene in these experiments. We additionally wanted to determine the level of endogenous expression of the glycosylase repair genes in the pulmonary epithelial A549 cell line. Others have demonstrated how different stressors lead to alterations in the endogenous production of specific repair genes. For example, it has been shown that endogenous gene expression of hOgg1was elevated following exposure to crocidolite asbestos which is known to cause an increase in 8-oxoG levels \[[@B48]\]. It has also previously been reported that treatment of A549 cells with sodium dichromate, a pro-oxidant, leads to a reduction of hOgg1protein expression that was not observed with H~2~O~2~\[[@B49]\]. One additional study demonstrated a dose dependent down regulation of hOgg1protein expression in rat lung after exposure to cadmium, a known carcinogen associated with the formation of intracellular ROS \[[@B50]\]. In our experiments we were able to demonstrate that both hyperoxia and IR do not appear to impact the endogenous expression of hOgg1, hMYH, and hMTHat 24 and 48 hours following exposure. It was noted that endogenous hNTHwas reduced after hyperoxia at 24 and 48 hours after the onset of exposure. One would speculate that this reduction in endogenous hNTHsecondary to hyperoxia is related to either decreased production or increased destruction in response to O~2~exposure. Over-expression of this repair enzyme did not result in improvements in survival after O~2~exposure based on our experiments. It may be that endogenous levels are adequate to correct this specific mutational burden for these experiments. Furthermore, no previous studies have determined how cells over-expressing specific repair genes may impact endogenous expression of the other oxidative BER genes under both normal and oxidative stress conditions. We were also able to demonstrate that endogenous expression of glycosylase repair genes were not altered under these conditions secondary to the over-expression of any of these genes. This is an important finding for interpretation of survival data; protection of cells is due to the over-expression of the specific gene and not due to enhancement of other endogenous repair enzyme levels, at least for the genes studied under these conditions. Some limitations may exist in using a lung carcinoma cell-line, which likely differs both in proliferative properties as well as in response to oxidative stress in comparison to primary epithelial cells. The enhanced cell growth observed with cell lines may be more reflective of undifferentiated alveolar type II cells which are likely to replace terminally differentiated alveolar type I cells after injury/death due to oxidative stress. This may not be a true reflection of growth under non-toxic conditions when very little cell division is occurring. This is an inherent problem observed when comparing cell lines with primary cells and results need to be interpreted in a way that considers this. It is difficult to know how this will translate to pulmonary epithelial cells in vivoat this stage. It certainly would appear that the protection observed is modest in degree in this pulmonary epithelial cell line. Further experiments assessing the function of the repair enzyme hMYHin this model will be important to perform in order to delineate the findings of slowed growth under normal conditions and improved survivability under conditions of O~2~and IR. More research looking at the potential for combination therapy, including DNA repair mechanisms in conjunction with other antioxidant defense mechanisms may be another approach to enhancing cell survival, which may lead to better clinical outcomes. Alternatively, cell survival may not be the most important end point for hyperoxia studies. Given that 8-oxoG, if left unrepaired, leads to G:C to T:A transversions, there may be an increase in mutational burden by these cells that isn\'t reflected in cell survival. Further experiments studying the impact on mutation production is underway. Ultimately, experiments need to be done in animal models to determine the translation to in vivopulmonary cells. Conclusions =========== In summary, we have demonstrated that over-expression of the DNA glycosylase repair enzyme hMYHmay enhance survival of a pulmonary epithelial cell line after exposure to conditions of IR and hyperoxia. We have also demonstrated that over-expression of hMYHleads to a slowing of growth of A549 cells under non-toxic conditions, which may in part play a role in this enhancement of survival by providing a wider window of opportunity for repair of oxidized lesions to occur. Lastly, we demonstrated that over-expression does not lead to altered endogenous expression of these repair genes. As the understanding of DNA repair mechanisms continues to grow and the evolution of gene therapy takes place, more treatment options may be available in the clinical setting to help with many disease processes including the damaging effects of oxygen and its metabolites. List of abbreviations ===================== apurinic, AP; base excision repair, BER; Dulbecco\'s modified Eagle\'s medium, DMEM; deoxyribose phosphate, dRP; enhanced green fluorescent protein, EGFP; fetal bovine serum, FBS; hydrogen peroxide, H~2~O~2~; ionizing radiation, IR; internal ribosomal entry site, IRES; long terminal repeat, LTR; oxygen, O~2~; Sulforhodamine B, SRB; reactive oxygen species, ROS; region of interest, ROI; Tris-Borate-EDTA, TBE; tris-buffered saline Tween-20, TBST; 8-oxoguanine, GO and 8-oxoG Authors\' contributions ======================= TK conducted the majority of the research experiments, performed the statistical analysis, and drafted the manuscript. MR conducted some of the cell survival experiments and participated in the design of the study. YX and XC helped with production of the lesion specific DNA repair genes. MK conceived of the study, and participated in its design and coordination. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by National Institutes of Health grantsNS38506, ES05865, ES03456, and P30 DK49218 supporting MRK.TK was supported on a T32 pulmonary training grant NIH-NHLBI/T32 (46886). Additional support was from The RileyChildren\'s Foundation.We would also like to thank Dr. Sankar Mitra for the hNTHclones, Dr. Amanda McCullough for the hMYHclones, and Dr. Laura Haneline for allowing us to use her oxygen chamber. Lastly, the Pediatric Pulmonary section at Riley Hospital for Children has been an additional provider of support and information throughout this research project.	PubMed Central	2024-06-05T03:55:48.341510	2004-9-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521691/", "journal": "Respir Res. 2004 Sep 27; 5(1):16", "authors": [ { "first": "Ted M", "last": "Kremer" }, { "first": "Mikael L", "last": "Rinne" }, { "first": "Yi", "last": "Xu" }, { "first": "Xian Ming", "last": "Chen" }, { "first": "Mark R", "last": "Kelley" } ] }
PMC521692	Background ========== Pediatric health-related quality of life (HRQOL) is increasingly acknowledged as an important health outcome measure in clinical trials and health services research and evaluation \[[@B1],[@B2]\]. Additionally, in pediatric chronic health conditions, the impact of disease and treatment on family functioning is a salient concern given the essential role of the family in child adaptation to disease \[[@B3]-[@B5]\]. Within this context, the impact of pediatric chronic health conditions on the family has been conceptualized within a theoretical risk and resistance framework, in which parent adjustment and the family system as a whole have been identified at increased risk \[[@B6]\]. Although there are a number of well-developed generic measures of family functioning, such as the Family Environment Scale \[[@B7]\], instruments that specifically measure the impact of pediatric chronic health conditions on parent and family functioning are less common. The two most widely utilized family impact instruments are the Impact on Family Scale and the Child Health Questionnaire (CHQ). The Impact on Family Scale-Revised is a brief unidimensional instrument that measures one factor of general negative impact on the social and familial systems and has demonstrated good reliability and validity in the samples tested \[[@B8]\]. The CHQ, a well validated instrument which contains scales measuring child HRQOL \[[@B9]\], contains a scale measuring whether the child\'s health or behavior limited family activities or caused family conflict. The CHQ also contains two parent self-report scales which measure the impact of the child\'s health on parent worry or concern and limitations in meeting their own needs. Although these two well-developed measures existed when we conceptualized the PedsQL™ Family Impact Module, after an analysis of the items and scales of the existing instruments, we felt that a PedsQL™ Family Impact Module would make a significant contribution to the literature by creating a multidimensional instrument that could stand alone, or be easily integrated into the PedsQL™ Measurement Model \[[@B10]\]. The PedsQL™ Measurement Model includes not only generic health-related quality of life \[[@B11]-[@B13]\] and disease-specific measurement instruments \[[@B14]-[@B18]\], but also generic measures of fatigue \[[@B15],[@B19]\], healthcare satisfaction \[[@B20],[@B21]\] and evaluations of the healthcare built environment \[[@B21]\]. Thus, we envisioned a Family Impact Module that would contribute to the literature by identifying items and scales which were not redundant with existing instruments, and which would further enhance the measurement options available through the PedsQL™ Measurement Model. In this context, the PedsQL™ Family Impact Module was developed and initially field tested in families with medically fragile children with complex chronic medical conditions as part of our evaluation of the healing environment of a Children\'s Convalescent Hospital \[[@B21]\]. In order to provide a contrast group to these children in this long-term care facility, we selected a population of children with comparable complex chronic medical conditions who were residing at home with their families. Since these children\'s severe medical conditions prevented them from providing self-report, the PedsQL™ Family Impact Module was designed as a parent proxy-report instrument. This study investigates the preliminary reliability and validity of the PedsQL™ Family Impact Module in medically fragile children with complex chronic health conditions. We hypothesized that the PedsQL™ Family Impact Module would distinguish between families in which the child resided at home versus those whose child resided in a long-term care facility based on the extant literature on pediatric chronic health conditions and the impact on parents and families \[[@B5],[@B6]\]. Method ====== Participants and Settings ------------------------- Participants were the parents of 23 medically fragile pediatric patients with complex chronic health conditions, such as severe cerebral palsy and birth defects. Participants from the Children\'s Convalescent Hospital (CCH) were parents of 12 pediatric patients who were residents of this long-term care facility. For each CCH family, the family member who completed the PedsQL™ Family Impact Module was the resident\'s mother. Participants from the REACH program (an outpatient program designed to reach out to families who choose to take care of their medically fragile children at home) were the parents of 11 pediatric patients. For each REACH family except one, the family member who completed the PedsQL™ was the patient\'s mother. PedsQL™ Family Impact Module ---------------------------- The 36-item PedsQL™ Family Impact Module Scales encompass 6 scales measuring parent self-reported functioning: 1) Physical Functioning (6 items), 2) Emotional Functioning (5 items), 3) Social Functioning (4 items), 4) Cognitive Functioning (5 items), 5) Communication (3 items), 6) Worry (5 items), and 2 scales measuring parent-reported family functioning; 7) Daily Activities (3 items) and 8) Family Relationships (5 items). Items and scales were developed through focus groups, cognitive interviews and pre-testing measurement development protocols \[[@B10],[@B11]\], and our prior research and clinical experiences with children with chronic health conditions and their families. Table [1](#T1){ref-type="table"} contains a general description of the scale items. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### PedsQL™ Family Impact Module -- general content of scales ::: Parent Functioning \# Items General Content ------------------------ -------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Physical Functioning 6 Problems with physical functioning, including feeling tired, getting headaches, feeling weak, and stomach problems Emotional Functioning 5 Problems with emotional functioning, including anxiety, sadness, anger, frustration, and feeling helpless or hopeless Social Functioning 4 Problems with social functioning, including feeling isolated, difficulty getting support from others, and finding time or energy for social activities Cognitive Functioning 5 Problems with cognitive functioning, including difficulty maintaining attention, remembering things, and thinking quickly Communication 3 Problems with communication, including others not understanding the family\'s situation, difficulty talking about child\'s health condition, and communicating with health professionals Worry 5 Problems with worrying, including worrying about child\'s treatments and side effects, about others\' reactions to child\'s condition, about the effect of the illness on the rest of the family, and about child\'s future Family Functioning \# Items General Content Daily Activities 3 Problems with daily activities, including activities taking more time and effort, difficulty finding time and energy to finish household tasks Family Relationships 5 Problems with family relationships, including communication, stress, and conflicts between family members, and difficulty making decisions and solving problems as a family Total Score is computed by averaging all 36 items. Parent HRQOL Summary Score is computed by averaging 20 items in Physical, Emotional, Social, and Cognitive Functioning. Family Summary Score is computed by averaging 8 items in Daily Activities and Family Relationships. ::: The PedsQL™ Family Impact Module was developed as a parent-report instrument. A 5-point response scale is utilized (0 = never a problem; 4 = always a problem). Items are reverse-scored and linearly transformed to a 0--100 scale (0 = 100, 1 = 75, 2 = 50, 3 = 25, 4 = 0), so that higher scores indicate better functioning (less negative impact). Scale Scores are computed as the sum of the items divided by the number of items answered (this accounts for missing data). If more than 50% of the items in the scale are missing, the Scale Score is not computed \[[@B22]\]. Although there are other strategies for imputing missing values, this computation is consistent with the previous PedsQL™ peer-reviewed publications, as well as other well-established HRQOL measures \[[@B23],[@B24]\]. The PedsQL Family Impact Module Total Scale Score is the sum of all 36 items divided by the number of items answered. The Parent HRQOL Summary Score (20 items) is computed as the sum of the items divided by the number of items answered in the Physical, Emotional, Social, and Cognitive Functioning Scales. The Family Functioning Summary Score (8 items) is computed as the sum of the items divided by the number of items answered in the Daily Activities and Family Relationships Scales. Procedure --------- The PedsQL™ Family Impact Module was mailed to families whose children were residents at the CCH and outpatients in the REACH program, along with a self-addressed stamped envelope in which to return the survey to the research team. A letter was included in the packet explaining the study, the confidentiality with which their data would be treated, and that the healthcare staff would not see this information. The protocol was approved by the Institutional Review Board at Children\'s Hospital and Health Center, San Diego. Statistical Analysis -------------------- Scale internal consistency reliability was determined by calculating Cronbach\'s coefficient alpha \[[@B25]\]. Scales with reliabilities of 0.70 or greater are recommended for comparing patient groups, while a reliability criterion of 0.90 is recommended for analyzing individual patient scale scores \[[@B26],[@B27]\]. Construct validity for the PedsQL™ Family Impact Module was determined utilizing the known-groups method. The known-groups method compares scale scores across groups known or expected to differ in the construct being investigated. In this study, PedsQL™ Family Impact Module scores in groups differing in residence of the child (Convalescent Hospital inpatient sample versus REACH outpatient sample) were computed \[[@B28],[@B29]\], using independent sample t-tests. We hypothesized that families whose children were residents in the Convalescent Hospital would report significantly higher scores (less negative impact) than families whose children were being taken care of at home based on the exant literature on pediatric chronic health conditions, families, and parental adjustment \[[@B6]\]. In order to determine the magnitude of the differences between families, effect sizes were calculated \[[@B30]\]. Effect size as utilized in these analyses was calculated by taking the difference between the Convalescent Hospital sample mean and the REACH sample mean, divided by the pooled standard deviation. Effect sizes for differences in means are designated as small (.20), medium (.50), and large (.80) in magnitude \[[@B30]\]. Statistical analyses were conducted using SPSS for Windows. Results ======= Means and Standard Deviations ----------------------------- Table [2](#T2){ref-type="table"} presents the means and standard deviations of the Convalescent Hospital inpatient sample and the REACH outpatient sample. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Scale descriptives for PedsQL™ Family Impact Module: Comparisons across CCH and REACH samples ::: [CCH Sample]{.underline} [REACH Sample]{.underline} ----------------------- -------------- ------------------------------ -------------------------------- -------- ------- ---------- -------- ---------------- ----------------- Scale \# Items N Mean SD N Mean SD Difference Effect Size Total Impact Score 36 12 81.00 17.06 11 62.49 17.26 18.51\\ 1.08 Parent HRQOL Summary 20 12 83.75 15.55 11 62.94 19.83 20.81\\\* 1.17 Physical Functioning 6 12 82.99 17.36 11 53.03 22.83 29.26\\\* 1.45 Emotional Functioning 5 12 78.33 18.26 11 64.48 26.59 13.85 0.61 Social Functioning 4 12 85.42 17.34 11 61.93 25.99 23.49\\ 1.07 Cognitive Functioning 5 12 88.75 12.81 11 74.09 18.95 14.66\* 0.91 Communication 3 12 73.61 24.58 11 52.15 24.67 21.46\* 0.87 Worry 5 12 69.17 21.09 11 56.82 25.52 12.35 0.53 Family Summary 8 12 84.27 20.47 11 68.81 24.11 15.46 0.69 Daily Activities 3 12 85.14 24.75 11 51.89 31.48 33.25\\\* 1.18 Family Relationships 5 12 83.75 23.07 11 78.95 27.62 4.80 0.19 Note: Higher values equal better health-related quality of life and family functioning. HRQOL = health-related quality of life; CCH = Children\'s Convalescent Hospital. REACH = outpatient sample. \p \< .05, \\p \< .02, \\\p \< .01; equal variances not assumed. Effect sizes are designated as small (.20), medium (.50), and large (.80). ::: Internal Consistency Reliability -------------------------------- Internal consistency reliability alpha coefficients for the PedsQL™ Family Impact Module Scales are presented in Table [3](#T3){ref-type="table"}. The scales exceeded the minimum reliability standard of 0.70 \[[@B26]\]. Most PedsQL™ Family Impact Module Scales approached or exceeded the reliability criterion of 0.90 recommended for analyzing individual patient scale scores \[[@B26],[@B27]\]. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### PedsQL™ Family Impact Module: Internal consistency reliability for total, CCH, and REACH samples ::: Scale Total N CCH N REACH N ---------------------------- ----------- ------- --------- ------- ----------- ------- Total Impact Score .97 23 .97 12 .95 11 Parent HRQOL Summary Score .96 23 .96 12 .95 11 Physical Functioning .91 23 .84 12 .88 11 Emotional Functioning .90 23 .83 12 .93 11 Social Functioning .88 23 .87 12 .88 11 Cognitive Functioning .93 23 .93 12 .91 11 Communication .88 23 .79 12 .95 11 Worry .82 23 .80 12 .84 11 Family Summary Score .90 23 .93 12 .89 11 Daily Activities .91 23 .95 12 .83 11 Family Relationships .97 23 .98 12 .96 11 Note: HRQOL = health-related quality of life; CCH = Children\'s Convalescent Hospital. REACH = outpatient sample. ::: Construct Validity ------------------ Table [2](#T2){ref-type="table"} presents the effect sizes and t-test results of the PedsQL™ Family Impact Module Scales for families with children at the CCH and REACH. The effects sizes were all in the medium to large effect size range except for one scale. Although the small sample size decreases the probability of detecting statistically significant differences, 7 of the 11 comparisons were statistically significant. Discussion ========== This study presents the preliminary reliability and validity of the newly developed PedsQL™ Family Impact Module. All internal consistency reliabilities exceeded the recommended minimum alpha coefficient standard of 0.70 for group comparisons, with most scales approaching or exceeding an alpha of 0.90, recommended for individual patient analysis \[[@B26]\]. The PedsQL™ Family Impact Module Scales performed as hypothesized utilizing the known-groups method. Where statistically significant differences existed between families with children at the CCH and REACH, REACH families were lower functioning, generally confirming the hypothesis that families whose medically fragile children live in a residential facility are higher functioning than those whose children live in the home. The present findings have certain limitations. Information on nonparticipants and an accurate response rate were not available, which may limit the generalizability of the findings. The generalizability of the findings is further limited by the small sample size and the selection of medically fragile children with complex chronic medical conditions. Whether the instrument would perform well in groups of children with other chronic health conditions is a matter of empirical inquiry. Given that instrument validation is an iterative process and consistent with this paradigm, the PedsQL™ Family Impact Module will be further field tested in other pediatric chronic health conditions with larger populations of children. Conclusion ========== The study demonstrates the preliminary reliability and validity of the PedsQL™ Family Impact Module, an instrument designed to assess the impact of pediatric chronic health conditions on parents\' HRQOL and family functioning. As predicted, families of children with medically fragile conditions who resided in a children\'s convalescent hospital were higher functioning than families of similar children who resided at home. List of Abbreviations ===================== HRQOL Health-Related Quality of Life PedsQL™ Pediatric Quality of Life Inventory™ Authors\' Contributions ======================= JWV and PD conceptualized the rationale and design of the study. JWV designed the instrument and drafted the manuscript. PED participated in the instrument design and coordination of initial data collection. SAS performed the statistical analysis and participated in study coordination, instrument development, and data collection. TMB participated in study conceptualization and design, instrument development, and data collection. All authors read and approved the final manuscript. Acknowledgements ================ This research was supported in part by a grant from the March of Dimes Foundation. The PedsQL™ Family Impact Module is available at <http://www.pedsql.org>.	PubMed Central	2024-06-05T03:55:48.344648	2004-9-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521692/", "journal": "Health Qual Life Outcomes. 2004 Sep 27; 2:55", "authors": [ { "first": "James W", "last": "Varni" }, { "first": "Sandra A", "last": "Sherman" }, { "first": "Tasha M", "last": "Burwinkle" }, { "first": "Paige E", "last": "Dickinson" }, { "first": "Pamela", "last": "Dixon" } ] }
PMC521693	Background ========== The potential impact of missing survey responses is often ignored in health-related quality of life (HRQL) studies \[[@B1],[@B2]\]. Missing data from study participants can cause bias in parameter estimates of models predicting HRQL outcomes \[[@B2]-[@B4]\]. Unfortunately, regression models are frequently interpreted with the assumption that available data are representative of the entire study population. Researchers may compare clinical characteristics of respondents with and without missing surveys, but rarely attempt to assess the impact of these differences on the regression model parameter estimates and ultimately on the results of the study. The assumption that there are minimal or no effects on parameter estimates is only reasonable if one can demonstrate that data missing from a study are truly missing at random, making them ignorable, which is rarely the case \[[@B2],[@B4]\]. Newer statistical techniques have been developed to assess and correct for bias resulting from missing HRQL surveys \[[@B2],[@B3]\]. One technique which has received little attention in the medical literature to date is the Heckman two-step method \[[@B5]-[@B8]\]. The Heckman method was developed by an economist, James Heckman, to address problems of self-selection among women participating in the labor force. This method makes it possible to assess whether selection bias is present, identify factors contributing to the selection bias, and to control for this bias in estimating the outcomes of interest. The Heckman method attempts to control for the effect of non-random selection by incorporating both the observed and unobserved factors that affect non-response. The objective of this study was to demonstrate the use of the Heckman two-step method to assess and correct for bias due to missing HRQL surveys. To accomplish this goal, we evaluated HRQL outcomes in a cohort of patients with acute coronary syndromes (ACS). Methods ======= Study population ---------------- We analyzed data from the VA Access To Cardiology study, which was a multi-center prospective cohort study of 2,733 veterans with a confirmed diagnosis of acute coronary syndromes (ACS), including either acute myocardial infarction or unstable angina \[[@B9]\]. Baseline patient characteristics (demographic, cardiac history, non-cardiac history and hospitalization variables) were collected at the time of ACS hospitalization. HRQL outcomes were then assessed by the Short-Form 36 (SF-36) health status survey which was mailed to all patients who were alive 7 months following ACS discharge. A second mailed survey was sent to non-respondents. If no response was obtained from the mailed surveys, attempts were made to contact the patients by phone. Of the 2,733 patients in the study, 1,660 (61%) completed the survey, 306 (11%) died, and 767 (28%) were alive and did not complete the survey. Of those completing the survey, most responded to the first mailing with much smaller numbers responding to the second mailing or to phone calls. Variables --------- The outcome variables were the Physical Component Summary (PCS) and Mental Component Summary (MCS) scores from the 7-month SF-36 health status survey. The PCS and MCS scores reflect a patient\'s overall physical and mental health status, respectively \[[@B10],[@B11]\]. The PCS and MCS scores are continuous variables with a range of 0--100, where higher scores indicate better health status. We constructed a dichotomous variable to indicate whether the patient responded to the SF-36 or not. It is important to note that models of HRQL outcomes may be biased both because subjects may have died before survey administration (survivor bias) or because of survey non-response in subjects that were alive at the time of survey administration \[[@B2]\]. However, the best way to handle patients who die prior to administration of the HRQL survey remains controversial \[[@B12]\]. Since the focus of this paper was to demonstrate the use of the Heckman model rather than methods of dealing with death in HRQL studies, we included only those patients who survived 7-months and were therefore eligible to complete the survey. Candidate predictor variables included a wide array of demographic, cardiac, and non-cardiac variables from the index hospitalization, and selected variables from the interim period between discharge and the 7-month SF-36 health status survey (Table [1](#T1){ref-type="table"}). These variables were derived from the established literature on risk factors for adverse post-MI outcomes (mortality, functional status, and HRQL) \[[@B13]-[@B20]\]. Patient demographic and clinical data from the index hospitalization and 7-month follow-up period were abstracted from the electronic medical record and/or from national VA patient care databases. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Comparison of SF-36 Survey Responders to Non-Responders ::: Variables SF 36 responders SF 36 non-responders P-value -------------------------------------------------------- ---------------------- -------------------------- ------------- *Demographics* Mean patient age 65.5 62.8 \<.01 Male gender 97.7% 97.5% 0.69 Caucasian race 78.4% 80.0% 0.21 *Prior Cardiac History* Hx of MI 38.0% 36.6% 0.33 Hx of Chronic heart failure 18.2% 18.3% 0.90 Hx of Coronary artery bypass graft surgery 26.3% 23.0% 0.01 Hx of Percutaneous coronary intervention 17.2% 12.6% \<.01 *Prior Non-Cardiac History* History of trauma 0.4% 0.7% 0.31 Substance or alcohol abuse 15.6% 21.0% \<.01 Current smoker 35.7% 43.0% \<.01 Dementia 2.0% 3.3% \<.01 COPD 29.7% 26.1% 0.01 Stroke 10.8% 13.2% 0.01 Depression 28.7% 40.5% \<.01 Arthritis 54.8% 54.4% 0.92 Diabetes mellitus 31.6% 32.0% 0.73 Peptic ulcer disease 11.1% 9.6% 0.26 Events during Index Admission and Interim Period\* Coronary revascularization during index admission 26.6% 23.9% 0.05 Cardiogenic shock 1.1% 1.1% 0.94 Hypotensive episode 8.7% 9.2% 0.59 Do not resuscitate order 1.2% 2.9% \<.01 Positive stress test 12.1% 11.1% 0.32 Admitted to Tertiary VA 63.9% 69.5% 0.01 ST-segment elevation MI on ECG 19.0% 21.4% 0.05 Mean serum creatinine 1.17 1.19 0.58 Left bundle branch block on ECG 3.6% 3.2% 0.52 Cardiac catheterization during index admission 33.4% 31.4% 0.16 Discharge diagnosis Unstable Angina (vs. MI) 49.4% 46.5% 0.06 \* Interim period is defined as the time period between discharge from index ACS hospitalization and the 7-month HRQL survey assessment. MI: myocardial infarction; Hx: History; COPD: chronic obstructive pulmonary disease; ECG: electrocardiogram ::: Analyses -------- Baseline characteristics of the patients who did and did not complete an SF-36 were compared using analysis of variance for continuous variables and chi-square for categorical variables (Table [1](#T1){ref-type="table"}). Then, a series of regression models were developed, including 1) initial PCS and MCS models which did not account for potential bias due to missing surveys, 2) Heckman selection models (modeling response to the SF-36), and finally 3) final PCS and MCS models (accounting for potential bias due to missing surveys). We used robust regression for all equations (Stata version 8.0 SE), controlling for cluster sampling by VA medical center. In prior analyses, we established that the intra-class correlation, the measure of the effect of clustering by medical center in this case, was significant. As a result, it was necessary to control for bias due to autocorrelation, or similarity among patients within a medical center, compared to patients at a different medical center. Stata uses the Huber-White estimator to control for the bias due to clustering. This technique deflates the standard errors of the parameter estimates, in this case the coefficients, correcting the inference statistics. Overview of the Heckman method ------------------------------ There are two steps in the Heckman method. The first step is the development of a selection equation (i.e. a model of factors associated with survey non-response). This step includes derivation of a variable from the selection equation called the Inverse Mills Ratio (IMR). The second step of the Heckman method is the insertion of the IMR variable into the initial regression models (e.g. those not accounting for potential bias due to missing surveys) from a given study in order to assess for, and attempt to control for, selection bias. Heckman method: Step one ------------------------ The first step in the Heckman method is to create the selection model, which estimates whether or not the quality of life survey was completed. The Heckman selection equation is usually estimated using a probit estimator \[[@B5],[@B21]\]. The probit estimator requires a binary outcome variable, in this case whether the patient responded to the SF-36 or not (coded 1 for responder, 0 for non-responder). The candidate predictor variables for the selection model were those listed in Table [1](#T1){ref-type="table"}. Although the Heckman selection equation will usually have multiple variables, some of which will be the same variables that enter the multivariable models of HRQL outcomes, it is important that the Heckman selection equation contain at least one variable that can legitimately be excluded from the initial models to safeguard against colinearity between the Heckman selection equation and the initial regression models. This means that this variable (or set of variables) is, in theory, a factor influencing whether someone responded to the questionnaire, but not a factor in predicting their component scores on the SF-36. This variable or set of variables is called an instrument in econometrics, and should be a strong predictor of response in the selection equation. We therefore stress that it is essential that the candidate variables considered for the Heckman selection equation be as comprehensive as possible, not omitting any variables that may contribute to whether a person responds to the survey. Once the Heckman selection equation is estimated, the residuals (error term) from this equation are used to form a new variable called the Inverse Mills Ratio (IMR). The formula to create the IMR variable depends on the distributional assumption in the outcome equation. In most HRQL applications, the quality of life score is the outcome of interest and is usually estimated using multivariable linear regression. In this case, the distributional assumption of the error term is the standard normal distribution, so that the ratio of the standard normal probability density function (pdf) and cumulative density function (cdf) applied to the residuals for each individual in the data set is created. The ratio of pdf/cdf is the IMR. Each individual in the study sample receives an individual value of the IMR based on the residual observed for that individual in the selection equation. In this study, the value of the IMR for each individual represented the predicted probability that they completed the 7-month SF-36 survey. It is important to note that the IMR is a function not only of observed or measured variables that are included in the selection equation, but also of unobserved or unmeasured variables. These are captured through the error term or residual in the selection equation, and included through the non-linear function used to estimate the IMR. As a result, adding the IMR into the outcome equation introduces a term that attempts to capture both observed and unobserved variables that affect selection, or non-response. We estimated the Heckman model using the maximum likelihood estimation method in Stata version 8.0. In this approach, the outcome and selection models are estimated jointly, which can result in slightly different selection models for different outcomes, in this case the PCS and MCS scores from the SF-36. However, for clarity of presentation of the Heckman process, we present only one table of selection equation results (the probit estimation of the probability of returning the SF36), assuming that the patient survived to the 7-month survey point. Heckman method: Step two ------------------------ The second step of the Heckman method is to include the IMR as a separate predictor variable in the initial regression models. In this study, the IMR variable derived from our Heckman selection model was inserted into the initial PCS and MCS models. Once this variable is inserted, two factors can be evaluated to help determine whether there is significant bias from missing responses in the initial models. First, one can examine the significance of the IMR variable itself. If significant, it suggests there was significant bias in the initial model. However, one potential limitation of the Heckman method is that if the Heckman selection model is not well-specified, and the variables in the selection model do not predict response/non-response well, the IMR may be weaker than expected and the Heckman method may have limited power to detect bias. Therefore, a second factor to examine following the addition of the IMR variable into the initial outcome models is whether or not there have been significant changes in any of the parameter estimates of the other predictor variables in the model. While somewhat arbitrary, changes in parameter estimates of \>10% may indicate that these estimates were biased due to missing surveys. Where possible, one should apply clinical judgment about whether changes in parameter estimates are \'biologically important\' \[[@B22]\]. With these factors taken together, the insertion of the IMR variable into the initial risk models allows the assessment of whether or not there was bias in the initial models, and suggests which initial predictors may have been most associated with this bias. Furthermore, by including the effect of unmeasured as well as measured variables from the selection equation, bias due to selection is controlled. Results ======= Baseline characteristics ------------------------ Compared to patients that completed the 7-month SF-36 survey, patients who did not respond to the survey were older, more likely to be current smokers and more likely to have a history of alcohol or substance abuse, dementia, stroke, or depression (Table [1](#T1){ref-type="table"}). Furthermore, the non-responders were more likely to have had ST-segment elevation on their ECG, more likely to have been admitted to a tertiary care VA hospital, and were more likely to have had a do not resuscitate order during their index hospitalization. Survey non-responders were less likely to have a history of prior coronary artery bypass graft (CABG) surgery, prior percutaneous coronary intervention (PCI), or chronic obstructive pulmonary disease (COPD), and were less likely to receive coronary revascularization during index hospitalization. Initial PCS and MCS models -------------------------- The initial multivariable PCS and MCS models not accounting for potential selection bias from missing HRQL surveys are presented in Tables [3](#T3){ref-type="table"} and [5](#T5){ref-type="table"}. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Heckman Selection Model (N = 1605) ::: 95% Confidence Interval --------------------------------------------------- ----------- ----------------------------- ----------- Mean patient age 0.01 0.00 0.02 Male gender 0.44 -0.07 0.95 Caucasian race -0.06 -0.25 0.13 Hx of MI 0.02 -0.16 0.20 Hx of Chronic heart failure -0.01 -0.20 0.18 Hx of Coronary artery bypass graft surgery 0.09 -0.13 0.31 Hx of Percutaneous coronary intervention 0.28 0.01 0.54 History of trauma\\ -0.29 -1.37 0.80 Substance or alcohol abuse -0.21 -0.40 -0.01 Current smoker -0.10 -0.29 0.09 Dementia 0.02 -0.76 0.79 COPD 0.23 0.05 0.41 Stroke -0.12 -0.39 0.15 Depression -0.26 -0.48 -0.05 Arthritis 0.08 -0.07 0.23 Diabetes mellitus -0.02 -0.25 0.21 Peptic ulcer disease\\ 0.04 -0.20 0.28 Coronary revascularization during index admission 0.12 -0.06 0.30 Cardiogenic shock 0.04 -0.81 0.89 Hypotensive episode 0.05 -0.29 0.38 Do not resuscitate order\\ -0.68 -1.09 -0.27 Positive stress test -0.01 -0.25 0.24 Admitted to Tertiary VA -0.09 -0.28 0.10 ST segment elevation MI on ECG -0.04 -0.29 0.22 Mean serum creatinine 0.01 -0.06 0.07 Left bundle branch block on ECG -0.02 -0.41 0.38 Cardiac catheterization during index admission 0.04 -0.10 0.19 Discharge diagnosis Unstable Angina (vs. MI) 0.06 -0.20 0.32 MI: myocardial infarction; Hx: History; COPD: chronic obstructive pulmonary disease; ECG: electrocardiogram; \\Instruments not included in outcomes equations ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Initial PCS Model ::: 95% Confidence Interval --------------------------------------------------- ----------- ----------------------------- ----------- Mean patient age (per 1 year increment) -0.06 -0.11 -0.01 Male gender -1.04 -4.31 2.23 Caucasian race -0.69 -1.72 0.35 Hx of MI -0.85 -2.46 0.75 Hx of Chronic heart failure -2.96 -4.36 -1.55 Hx of Coronary artery bypass graft surgery -3.12 -4.29 -1.95 Hx of Percutaneous coronary intervention -1.09 -2.68 0.51 Substance or alcohol abuse -0.30 -1.89 1.30 Current smoker -0.57 -1.66 0.52 Dementia -0.84 -3.17 1.50 COPD -3.76 -4.77 -2.76 Stroke -3.01 -4.34 -1.67 Depression -4.61 -6.07 -3.15 Arthritis -1.89 -2.94 -0.83 Diabetes mellitus -2.33 -3.56 -1.09 Coronary revascularization during index admission 1.91 0.38 3.44 Cardiogenic shock 1.62 -3.88 7.12 Hypotensive episode 0.13 -1.93 2.19 Positive stress test 1.15 -0.47 2.76 Admitted to Tertiary VA -0.88 -2.00 0.24 ST-segment elevation MI on ECG 2.34 0.81 3.86 Mean serum creatinine (per 1 mg/dl increment) -0.64 -1.13 -0.16 Left bundle branch block on ECG 0.09 -2.43 2.61 Cardiac catheterization during index admission 0.63 -0.45 1.71 Discharge diagnosis Unstable Angina (vs. MI) -0.66 -1.54 0.21 Inverse Mills Ratio \* \* Inverse Mills Ratio: Variable derived from the Heckman Selection Equation; MI: myocardial infarction; Hx: History; COPD: chronic obstructive pulmonary disease; ECG: electrocardiogram ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### PCS Model Corrected For Response Bias (N = 1605) ::: 95% Confidence Interval --------------------------------------------------- ----------- ----------------------------- ----------- Mean patient age (per 1 year increment) -0.07 -0.11 -0.03 Male gender -1.00 -4.05 2.04 Caucasian race -0.67 -1.62 0.28 Hx of MI -0.89 -2.39 0.62 Hx of Chronic heart failure -2.95 -4.22 -1.69 Hx of Coronary artery bypass graft surgery -3.16 -4.23 -2.09 Hx of Percutaneous coronary intervention -1.24 -2.75 0.28 Substance or alcohol abuse -0.17 -1.85 1.52 Current smoker -0.48 -1.48 0.51 Dementia -0.77 -2.89 1.34 COPD -3.91 -4.91 -2.91 Stroke -2.92 -4.14 -1.70 Depression -4.68 -6.06 -3.31 Arthritis -1.71 -2.75 -0.67 Diabetes mellitus -2.33 -3.45 -1.20 Coronary revascularization during index admission 1.76 0.32 3.19 Cardiogenic shock 1.44 -3.72 6.60 Hypotensive episode 0.48 -1.50 2.46 Positive stress test 1.06 -0.38 2.50 Admitted to Tertiary VA -0.78 -1.81 0.25 ST-segment elevation MI on ECG 2.41 0.97 3.85 Mean serum creatinine (per 1 mg/dl increment) -0.66 -1.11 -0.22 Left bundle branch block on ECG 0.05 -2.30 2.39 Cardiac catheterization during index admission 0.60 -0.41 1.60 Discharge diagnosis Unstable Angina (vs. MI) -0.62 -1.42 0.19 Inverse Mills Ratio \* -2.15 -5.66 1.37 \* Inverse Mills Ratio: Variable derived from the Heckman Selection Equation; MI: myocardial infarction; Hx: History; COPD: chronic obstructive pulmonary disease; ECG: electrocardiogram ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Initial MCS Model ::: 95% Confidence Interval --------------------------------------------------- ------------ ----------------------------- ----------- Mean patient age (per 1 year increment) 0.08 0.04 0.13 Male gender 0.14 -3.07 3.36 Caucasian race -0.69 -1.51 0.12 Hx of MI -1.19 -2.36 -0.02 Hx of Chronic heart failure 0.01 -1.05 1.07 Hx of Coronary artery bypass graft surgery 0.27 -0.94 1.48 Hx of Percutaneous coronary intervention 0.49 -0.76 1.75 Substance or alcohol abuse -2.27 -3.92 -0.62 Current smoker -0.51 -1.70 0.67 Dementia -2.80 -6.71 1.10 COPD -1.14 -2.15 -0.12 Stroke -2.10 -4.78 0.58 Depression -1.13 -2.56 0.30 Arthritis -11.68 -13.60 -9.76 Diabetes mellitus -0.49 -1.55 0.57 Coronary revascularization during index admission 1.09 -0.17 2.35 Cardiogenic shock -2.68 -8.31 2.95 Hypotensive episode -0.09 -2.52 2.33 Positive stress test -0.01 -2.44 2.43 Admitted to Tertiary VA 0.02 -1.37 1.40 ST-segment elevation MI on ECG 1.88 0.64 3.12 Mean serum creatinine (per 1 mg/dL increment) -0.34 -1.00 0.32 Left bundle branch block on ECG -1.87 -4.72 0.97 Cardiac catheterization during index admission -0.40 -1.62 0.81 Discharge diagnosis Unstable Angina (vs. MI) 0.24 -0.97 1.46 Inverse Mills Ratio \* \* Inverse Mills Ratio: Variable derived from the Heckman Selection Equation; MI: myocardial infarction; Hx: History; COPD: chronic obstructive pulmonary disease; ECG: electrocardiogram ::: Variables significantly associated with better 7-month physical health status included ST-segment elevation MI on electrocardiogram and coronary revascularization during the index ACS hospital admission. Variables significantly associated with worse 7-month physical health status included older age, history of prior CABG surgery, chronic heart failure, arthritis, COPD, stroke, depression, diabetes, and elevated serum creatinine during index ACS admission. Variables significantly associated with better 7-month mental health status included older age and ST-segment elevation MI on electrocardiogram. Variables significantly associated with worse 7-month mental health status included a history of prior MI, alcohol and/or substance abuse, COPD, and arthritis. Heckman selection model ----------------------- The Heckman selection model (modeling response to the SF-36) is presented in Table [2](#T2){ref-type="table"}. Older age, prior PCI, and history of COPD were associated with an increased likelihood of survey response, whereas a history of alcohol or substance abuse, depression, and have had a do not resuscitate order during their index hospitalization were associated with a decreased likelihood of survey response. Final PCS and MCS models ------------------------ The final multivariable models for PCS and MCS (after addition of the IMR variables from the Heckman selection model) are presented in Tables [4](#T4){ref-type="table"} and [6](#T6){ref-type="table"}. There was little evidence of selection bias for the PCS model. None of the results of inference testing for significance changed between the initial model and the model with the IMR variable added. Furthermore, the changes in magnitude of parameter estimates were not large, and the parameter estimate on the IMR variable itself was not significant. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### MCS Model Corrected For Response Bias (N = 1605) ::: 95% Confidence Interval --------------------------------------------------- ------------ ----------------------------- ----------- Mean patient age (per 1 year increment) 0.06 0.00 0.12 Male gender 0.25 -2.65 3.15 Caucasian race -0.51 -1.24 0.22 Hx of MI -1.24 -2.07 -0.40 Hx of Chronic heart failure 0.17 -0.94 1.27 Hx of Coronary artery bypass graft surgery -0.01 -1.21 1.18 Hx of Percutaneous coronary intervention -0.19 -1.36 0.99 Substance or alcohol abuse -1.55 -3.39 0.29 Current smoker -0.17 -1.46 1.13 Dementia -2.81 -6.90 1.27 COPD -1.76 -2.80 -0.72 Stroke -1.75 -4.25 0.75 Depression -1.33 -2.71 0.04 Arthritis -10.80 -12.83 -8.78 Diabetes mellitus -0.35 -1.43 0.72 Coronary revascularization during index admission 0.70 -0.48 1.89 Cardiogenic shock -2.58 -7.28 2.12 Hypotensive episode -0.16 -2.26 1.94 Positive stress test 0.01 -2.24 2.25 Admitted to Tertiary VA 0.24 -1.10 1.58 ST-segment elevation MI on ECG 1.92 0.91 2.92 Mean serum creatinine (per 1 mg/dL increment) -0.37 -0.97 0.24 Left bundle branch block on ECG -2.00 -4.63 0.63 Cardiac catheterization during index admission -0.46 -1.63 0.72 Discharge diagnosis Unstable Angina (vs. MI) 0.04 -1.23 1.32 Inverse Mills Ratio \* -8.93 -10.73 -7.13 \* Inverse Mills Ratio: Variable derived from the Heckman Selection Equation; MI: myocardial infarction; Hx: History; COPD: chronic obstructive pulmonary disease; ECG: electrocardiogram ::: By contrast, when the IMR variable was inserted into the initial MCS model, we found evidence of selection bias. In this case, the parameter estimate for history of alcohol or substance abuse changed from significant to insignificant with the introduction of the IMR variable. Therefore, it appears that a history of alcohol or substance abuse was associated with lower likelihood of responding to the survey, but not directly associated with mental health status. In addition, a number of parameter estimates changed quantitatively, with larger changes than those observed in the PCS findings. Finally, the coefficient on the IMR variable itself was significant, and was negatively associated with MCS, implying that unobserved variables in the selection equation appear to be associated with worse MCS scores. Discussion ========== The goal of this study was to demonstrate the use of the Heckman two-step method to assess and correct for bias due to missing HRQL surveys in a clinical study of acute coronary syndrome patients. We created initial multivariable models of 7-month post-ACS physical and mental health status using data only from survey respondents. Then, using the Heckman method, we modeled survey non-response, derived an Inverse Mills Ratio variable for each patient that captured the likelihood of survey response, and incorporated this variable into our initial models to assess and correct for potential bias from survey non-response. We found that our initial model of 7-month physical health status was not biased due to survey non-response. In contrast, our initial model of 7-month mental health status was biased. After controlling for bias due to non-response, a history of alcohol or substance abuse was no longer associated with mental health status. Furthermore, the magnitude of the parameter estimates for several of the other predictor variables in the MCS model changed after accounting for bias due to survey non-response. Given these results, biased parameter estimates of the association between these variables and mental health status would have been reported if we had used the standard approach to evaluating the predictors of HRQL outcomes in this population. Furthermore, we might have concluded that alcohol/substance abuse was significantly associated with mental health status outcomes following ACS, and may therefore be an important target for interventions to improve post-ACS HRQL (e.g. improving alcohol screening and treatment). While alcohol/substance abuse may be important for other reasons, it would have been incorrect to conclude that it was associated with HRQL in our study population. Rather, it was a marker for survey non-response. This analysis demonstrates the utility of the Heckman method in its application for assessing and correcting survey response bias in clinical studies of HRQL. Health-related quality of life data are usually not missing at random, and failure to account for missing HRQL assessments can bias estimates of associations and may lead to inappropriate conclusions about the determinants of HRQL outcomes \[[@B1]-[@B3]\]. Often, HRQL data are missing in systematic ways that can be estimated and controlled for. This study demonstrates the use of one technique to accomplish this, the Heckman two-step method \[[@B5]-[@B8]\]. To date, the Heckman method has rarely been utilized in studies reported in the medical literature, although it was previously used in one study assessing the impact of selection on medication use among older patients \[[@B7]\]. There are other statistical techniques, or approaches, to assess and correct for bias resulting from missing HRQL surveys, including index function models, propensity scores, instrumental variables, and multiple imputation methods \[[@B2]-[@B4],[@B23]\]. The Heckman method is one example of an index function model. Generally, the index function approach to missing HRQL data is to model whether or not HRQL surveys were completed (i.e. the dependent variable is survey completion). This allows an estimation of the \'likelihood\' that a given patient would complete a survey based on their clinical characteristics and/or other process or structure of care variables. This information, in turn, is used to assess and correct for bias in the primary model of interest (i.e. the model of quality of life outcome). Therefore, a primary strength of the Heckman method is that it not only permits the assessment of selection bias, it corrects for the bias, and does so in an informative way that may yield new insights into the association between patient characteristics or processes of care and outcomes of interest such as HRQL. In the Heckman method, the assumption is made that the error term in the outcome equation is standard normal, the distribution assumed in classical linear regression. Other index functions allow other distributional assumptions to be made for the error term in the outcome equation, such as logistic. Propensity score approaches can be analogous to the Heckman method in that a multivariable model of survey non-response is developed and the probability of survey non-response is used to stratify the study population and/or the propensity score is used as an independent variable in the primary HRQL models. In other words, propensity scores are similar to the Heckman method in that the predicted probability of non-response is used as the basis for assessing the impact of missing data and controlling for it \[[@B23]\]. Unlike a propensity score, however, which is often entered directly into the outcome equation as a predictor, the non-linear transformation from the prediction into the IMR variable in the Heckman method is one of the safeguards against colinearity in the outcome equation. Instrumental variable approaches are also used to address similar questions to those addressed by the Heckman method. In the full instrumental variable approach, a single exogenous variable (called the instrument) is used to stratify the full sample, removing the effect of the correlated error terms that lead to biased estimates \[[@B24]\]. An instrumental variable approach can be a very powerful approach to controlling selection bias. However, it can be very difficult to find an appropriate and adequate instrumental variable. The Heckman approach offers a more flexible, if less powerful, approach, and adds information about the underlying processes by which selection arose. It should be noted that propensity scores can be used as instrumental variables, when a suitable instrument is found. Finally, multiple imputation methods can be employed to address missing HRQL data \[[@B2]\]. In contrast to the Heckman and other approaches described thus far, multiple imputation methods derive missing values from existing data in the dataset, thereby creating a \'complete\' dataset and eliminating the need to drop patients from analysis. Imputation thereby eliminates bias from missing data per se (i.e. there is no longer missing data), but is highly dependent on the validity of deriving the missing HRQL survey data from the existing dataset. It is important to note that in this paper, we are focused on missing surveys rather than incomplete surveys (i.e. missing data elements within a survey). In this regard, multiple imputation will most often be employed in studies with serial measurements of HRQL over time, such that HRQL data before and/or after the time point of interest can inform the missing data imputation. The Heckman method can be used even for a single point in time, cross-sectional assessment of HRQL, as in our analysis in which we measured HRQL at only one time point. The Heckman method has several limitations. First, the selection equation must have at least one variable that is associated with survey response but not the outcome of the study (i.e. HRQL). In some clinical applications, the inability to identify such a variable may make it difficult to use the full Heckman method to control bias. However, it is still possible to use the first step of estimating a selection equation to assess the degree to which selection bias may affect the parameter estimates in an outcome equation. If there are variables that are significant in both the selection equation and the outcomes equation, it is likely that there is bias due to selection effects in the outcome equation. Acknowledging this and commenting on the likely magnitude of effect may provide helpful guidance to clinicians and other researchers. Another limitation of the Heckman method is that this technique depends heavily on the quality of the data available for the selection equation. If the amount of variance explained is relatively low, then there is a possibility that selection bias in the outcomes equation may not be detected. In other words, the Heckman method can be under-powered for the detection of bias in some cases. Finally, the Heckman method is very sensitive to how the model is specified; in other words, omitting variables that are associated with either non-response or with the outcome of interest (in this case, health related quality of life measures) can lead to inaccurate findings and biased estimates of the parameters in the final models. Careful attention to specifying the models, and ensuring that model specification follows what is known in the literature to be associated with the outcomes of interest is essential \[[@B4]\]. Conclusions =========== This study demonstrated the use of the Heckman two-step method to assess and control for bias from missing HRQL surveys in a clinical study. We found that our mental health status model was significantly biased due to missing HRQL assessments. Recognition and correction of this bias changed the parameter estimates of association and the factors that we concluded were associated with HRQL seven months following hospital admission for an acute coronary syndrome. We conclude that the Heckman two-step method may be a valuable tool in the assessment and correction of selection bias in clinical studies of HRQL. Authors\' contributions ======================= AES conceived of the study, participated in design and coordination, conducted analyses, and drafted the manuscript; MEP conducted analyses and contributed to the manuscript; DJM and JAS reviewed and contributed to the manuscript; JSR participated in the design and coordination of the study and participated in the drafting and revision of the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This study was supported by a grant from the Health Services Research and Development Service, Department of Veterans Affairs, ACC 97-079. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs. Dr Rumsfeld is supported by a VA Health Services Research and Development Advanced Research Career Development Award (ARCD 98341-2).	PubMed Central	2024-06-05T03:55:48.346603	2004-9-16	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521693/", "journal": "Health Qual Life Outcomes. 2004 Sep 16; 2:49", "authors": [ { "first": "Anne E", "last": "Sales" }, { "first": "Mary E", "last": "Plomondon" }, { "first": "David J", "last": "Magid" }, { "first": "John A", "last": "Spertus" }, { "first": "John S", "last": "Rumsfeld" } ] }
PMC521694	Background ========== Heart failure is among the most common chronic diseases in modern civilizations. The dilatation of the left ventricle frequently induces intracardiac conduction delays resulting in asynchronous left ventricular motion. This manifests as left bundle branch block in the surface ECG. Both QRS width and intraventricular asynchrony are predictors of hospitalization and severe cardiac events in patients with heart failure \[[@B1]-[@B3]\]. The mechanisms of myocardial asynchrony include a delayed left ventricular regional contraction and relaxation. The right ventricle contracts during left ventricular end-diastole, leading to a \"bulging\" of the septum into the left ventricle. The intra(left)ventricular delay of the systolic velocity induces the \"delayed longitudinal contraction (DLC)\". Furthermore, the delay of the contraction of the papillary muscles aggravates mitral regurgitation. This, in summary, leads to an increased oxygen demand of the myocardium \[[@B4]\]. Resynchronization of the intraventricular conduction can be achieved by introducing an additional lead through the coronary venous sinus to stimulate the left ventricle (biventricular pacing, BVP). The combination of BVP and a cardioverter-defibrillator (ICD) combines the clinical improvement by BVP and reduction in mortality \[[@B5]\]. Recent studies have shown an acute and sustained hemodynamic improvement, reversal of LV-remodeling, an increased quality of life, a reduction of symptoms of heart failure, and improvement of exercise tolerance after biventricular pacing. Markers of reverse remodeling were reduction of left ventricular volumes, increase in LVEF without an increase in oxygen consumption, reduction of mitral regurgitation \[[@B6]-[@B10]\]. However, a significant reduction of mortality after BVP alone could not be demonstrated. In the current guidelines, LBBB in the surface ECG and a reduced LVEF are the main indications for BVP \[[@B11]\]. However, about one third of patients in the large multicenter BVP studies did not improve -- despite BVP \[[@B6],[@B7],[@B12]\]. There is increasing evidence, that there is only a weak correlation of electrical (QRS width) and mechanical asynchrony and the benefit of BVP. It seems that not all heart failure patients with LBBB have mechanical asynchrony \[[@B12],[@B13]\]. Furthermore, asynchrony is common even in heart failure patients with narrow QRS complexes compared with healthy controls. A prospective study assessed left ventricular systolic and diastolic asynchrony in 67 patients with heart failure (LVEF \< 50%) with normal QRS width and 45 patients with CHF and wide QRS complexes (\>120 ms). 88 healthy control patients were included. Systolic (diastolic) asynchrony occurs in 51% (46%) of the heart failure patients with narrow QRS complexes and in 73% (69%) in the patients with wide QRS complexes. Systolic asynchrony was defined as the max difference in time-to-peak myocardial contraction of 12 myocardial segments. Diastolic asynchrony was defined as maximum difference of time-to-peak early diastolic relaxation. In summary, the authors state that asynchrony is common in patients with heart failure even without a wide QRS complex \[[@B14]\]. This is confirmed in a study with 158 heart failure patients (LVEF \< 35%), that were divided in three subgroups: Group 1 with no (QRS \< 120 ms), group 2 with mild (120--150 ms) and group 3 with severe LBBB (\>150 ms). Interventricular asynchrony was defined by TDI as IVMD \>40 ms and the intraventricular delay as maximum pre-ejection period of \>50 ms in one or more myocardial segments. Asynchrony was seen in all three subgroups, however, there was no correlation between interventricular and intraventricular asynchrony \[[@B15]\]. A recent study demonstrated that successful BVP can be achieved in patients with a normal QRS duration and asynchrony \[[@B16],[@B17]\]. These controversial data indicate the need for a more careful patient selection for BVP. Newer echocardiographic techniques, such as Tissue Doppler Imaging and Tissue Synchronization Imaging could potentially improve patient selection and guidance of implantation and programming of the devices for BVP. The risks of pacemaker implantation and expenses in non-responders to BVP could be avoided. Furthermore, the cost-effectiveness of BVP would be augmented. Definitions of asynchrony ========================= Regarding the nomenclature, the term \"asynchrony\" is used synonymously to \"dyssynchrony\" in this article. There is a variety of methods to determine asynchrony. In table [1](#T1){ref-type="table"}, the different approaches to asynchrony are listed concisely. The QRS width (LBBB \> 120, 130, 150 ms) is the simplest method, but the sensitivity to predict benefit from BVP is rather low \[[@B18],[@B19]\]. Magnetic resonance imaging can also detect areas of asynchrony but this technique can not be repeated for follow-up after device implantation. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Concise summary of the different approaches to echocardiographic measurement of asynchrony ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Assessment of asynchrony with: Ref. Criteria Segments Limitations Analysis time Prediction of benefit -------------------------------------------------- ------------ ---------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------- ------------------------------------------------------------------------ ----------------------------------- ----------------------------------------------------- I. Global ventricular asynchrony ECG 4, 44 QRS width \>120 ms Global assessment LBBB after myocardial infarction Short Low (30% non-responder) M-mode 21 Septal-to-posterior wall motion delay \>130 ms septal and posterior scar tissue, only septal or posterior Short low pw-TDI 25 Cumulative asynchrony (EMD) \>102 ms Intra LV (5 basal segments) and interventricular (vs. RV lateral segment) Low spatial resolution Long Good prediction of acute response (AUC in ROC 0,84) II. Interventricular asynchrony pw-Doppler echocardiography 47 Interventricular mechanical delay (IVMD) \>40 ms Aortic and pulmonary outflow tract Not simultaneous Short No III. Intraventricular asynchrony 2D-TDI 27 Ts-SD: intraventricular systolic asynchrony index: \>33 ms 12 segments complex (post-processing) Long Acute response (3 months) 4, 32 Difference in septal-lateral time-to-peak TDI, cut-off \>60 ms 12 segments Complex Long EF increase after BVP 40 Mean regional myocardial performance index: Difference between regional Q-wave-to.peak systolic displacement times 12 segments\ Complex Long, offline Acute response 4 segments 33 Ts-SD: cut-off: 31,4 ms 12 segments Complex Long 3 months response, reverse remodeling Strain and strain rate 15, 34, 33 Myocardial deformation in systole, presence of post-systolic shortening 12 segments Complex, time consuming, in dilated ventricles low spatial resolution) Long Controversial data Tissue Tracking 28 DLC in \>2 basal segments 12 basal segments in apical four chamber view. Requires correct timing of LV events Short Acute response TSI 36 Color-coded time-to-peak tissue Doppler velocities (cut-off \>65 ms in anteroseptum and posterior wall in apical long axis view) 16 segments except apex Only velocity data Short Acute response (Sensitivity 87% Specificity 100%) 3-D-echo 26 No quantitative criteria defined All segments Reduced spatial resolution Time consuming, off-line analysis No systematic data Automated endocardial border detection (ABD) 26 Septal-lateral phase angle difference 100 segments. apical-four-chamber view (septal-lateral) High complexity, single imaging plane Long, only off-line Acute response ABD + Contrast 46 Echo-contrast cardiac variability imaging: displacement maps apical four chamber High complexity, single imaging plane Long Acute response ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: Echocardiographic tools include 2D, Doppler and Tissue Doppler methods. Up to date, there is no consensus on the definition of echocardiographically measured myocardial asynchrony. The determination of asynchrony by M-mode echocardiography is limited to septal and inferior segments in parasternal long-axis and is not performed routinely in current studies \[[@B20],[@B21]\]. Earlier echocardiographic approaches to asynchrony included the delayed long axis shortening that was found to suppress early diastolic transmitral flow and subsequently leading to decreased leftventricular function \[[@B22]\]. Tissue Doppler imaging (TDI) measures regional wall motion velocities. TDI can accurately quantify regional left ventricular function \[[@B23]\]. Pulsed wave TDI does not allow simultaneous comparison of regional timing in different segments within one cardiac cycle. Color-coded TDI reduces beat-to-beat variability and examination time. Color coded TDI has a very high time resolution of 10 ms. TDI technology includes tissue tracking and strain rate imaging. Tissue tracking allows the measurement and visualization of longitudinal motion in each myocardial segment during the different phases of the heart cycle. Strain measures compression and distension of myocardial segments (\"deformation imaging\") and strain rate imaging expresses strain changes per time interval. Post-systolic movement diagnosed with velocity or tissue tracking can be differentiated into passive or active motion (=contraction, then defined as PSS). But in ischemic cardiomyopathy PSS was not an useful criterion for response to BVP because this phenomenon is not only a sign of asynchrony but also a marker for ischemia and/or viability of severe hypo/akinetic segments \[[@B24]\]. Tissue synchronization imaging (TSI) is a newer technique that utilizes color-coded time-to-peak tissue Doppler velocities and visualizes segments of dyssynchrony in real-time by superimposing these temporal motion data on 2D echo images. TSI analysis is possible in all myocardial regions except the apex. The color-coding is green (normal time-to-peak velocity: 20--150 ms), yellow (150--300 ms) and red (300--500 ms) \[[@B25]\]. Online 3D echocardiography and automated border detection (ABD) might be future diagnostic tools to diagnose asynchrony but need evaluation in larger studies \[[@B26]\]. Myocardial asynchrony includes inter- and intraventricular asynchrony. [Interventricular asynchrony]{.underline} can be assessed by comparing pw-Doppler signals in the right and left ventricular outflow tracts. A delay of \>60 ms is considered to demonstrate interventricular asynchrony. These measurements in the outflow tracts cannot be performed simultaneously and, therefore, there is a high inter-measurement variability and dependence on cardiac workload. In addition, interventricular asynchrony can measured as the difference of the electromechanical delays in the basal LV segments and the lateral RV segments \[[@B35]\]. [Intra(left)ventricular]{.underline} asynchrony is considered to be the most important aspect of the electromechanical delay (EMD). It can be measured by a variety of methods. EMD is defined as the delay between the onset of the QRS complex on the surface ECG and the onset of the systolic TDI wave in corresponding myocardial segments. Recently, the systolic synchronicity index has been introduced \[[@B27]\]. It is defined as the standard deviation (SD) of the EMD in 12 LV segments (6 basal, 6 mid-segmental model). Intraventricular asynchrony can also be demonstrated by tissue tracking with diastolic color-coded areas called DLC. This is the amount of post-systolic contraction after the closure of the aortic valve (i.e. post systolic shortening = PSS) which was confirmed by strain and strain rate in this study \[[@B28]\]. Intra-left ventricular asynchrony is not only of diagnostic value for selecting patients for BVP, but has prognostic value as well. Bader et al. \[[@B3]\] examined inter- and intraventricular asynchrony as an independent predictor of heart failure worsening: 104 patients with chronic stable heart failure without previous myocardial infarction (LVEF \< 45%) were included, follow-up echocardiography was performed after one year. Study endpoint of heart failure worsening was hospitalization for cardiac decompensation. Intra-left ventricular asynchrony is an independent predictor of severe cardiac events. Only a weak correlation of intra/inter-ventricular EMD and QRS width could be demonstrated. In figures [1](#F1){ref-type="fig"}, [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"}, [7](#F7){ref-type="fig"}, [8](#F8){ref-type="fig"}, [9](#F9){ref-type="fig"}, the different approaches to assess asynchrony as well as echocardiographic examples of successful biventricular pacing are illustrated. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Measurement of interventricular mechanical (IMD) delay by PW Doppler: A) PW Doppler in aortic outflow tract: Measurement from onset of QRS to the onset of PW curve in the aortic outflow tract. This time is also called the aortic pre-ejection time and is a marker for intra(left)ventricular asynchrony. B) PW Doppler in pulmonary outflow tract: Measurement from onset of QRS to the onset of PW curve in the pulmonary outflow tract. The IMD is the difference between the time of a) and b). ::: ![](1476-7120-2-17-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Assessment of asynchrony in parasternal long axis view by M-mode: Time difference between peak of septal and inferior myocardial contraction. ::: ![](1476-7120-2-17-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Tissue Doppler velocitydata for the quantification of asynchrony from apical four chamber view. Sample volumes are in the basal lateral and basal septal segment. A) Normal control patient. There is a synchronous myocardial velocity in the septal (=yellow) and the lateral (=green curve) segment. IVC = isovolumetric contraction, IVR = isovolumetric relaxation, S = peak systolic velocity; E = early diastolic filling, A = late (atrial) diastolic filling. B) There is asynchronous myocardial velocity in the septal (=yellow) and the lateral (=green curve) segment. ::: ![](1476-7120-2-17-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Assessment of asynchrony by strainfrom the apical four chamber view. The sample volumes are in the basal septal and the basal lateral segments. A) Normal strain curve in a control patient. ICT = isovolumetric contraction time. B) Strain curve with asynchronous myocardial velocity in the septal (=yellow) and the lateral (=green curve) segment. ::: ![](1476-7120-2-17-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Tissue Trackingallows the visualization of longitudinal motion in each myocardial segment. Images are from the apical four chamber view. A) Normal control patient. There are normal colour-coded displacement values in the lateral and septal segments, with physiologically higher values in the more basal segments and lower values towards the apex. B) Tissue Tracking in a patient with dilated cardiomyopathy. There is a dilated left ventricle with \"baseball shape\" and reduced displacement values and no basal-apical gradient (max displacement = 8 mm) in the septal segments and DLC in the lateral wall (no colour-coding) indicating asynchrony of the lateral wall. ::: ![](1476-7120-2-17-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Tissue Synchronization Imagingdisplays colour-coded time-to-peak tissue Doppler velocities. The colour-coding is green (normal time-to-peak velocity: 20--150 ms), yellow (150--300 ms) and red (300--500 ms) Apical four chamber view. A) TSI in a control patient (only green colour coding indicating synchronous contraction) B) TSI in a patient with LBBB: The basal and mid-septal segments show a delayed time-to-peak velocity (red colour). ::: ![](1476-7120-2-17-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Delayed longitudinal contraction (=DLC) as a marker for asynchrony can be visualized by analysis of systolic and diastolic Tissue Tracking. Systolic tracking analyzes the systolic displacement i.e. tracking interval between the onset of QRS-complex and the end of the T-wave. Diastolic tracking can demonstrate DLC with colour coding (end of T until begin of R). Images from apical two chamber view A) Systolic Tracking: The inferior segments (=grey area) show DLC with no systolic motion B) Diastolic Tracking: The inferior segments (=colour coded area) show DLC with diastolic movement. ::: ![](1476-7120-2-17-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Demonstration of successful BVP by Tissue Tracking in apical four chamber view in a patient with dilated cardiomyopathy. Images from apical four chamber view. A) Before BVP, there is a dilated ventricle (\"baseball shape\") with reduced systolic displacement (max displacement = 8 mm) in the septum and DLC in the lateral wall (no colour-coding) indicating asynchrony of the lateral wall. B) After three months of BVP, there is a reduction of left ventricular dilatation (reverse remodelling, \"American football shape\" of the left ventricle), increased tracking values and no DLC regions anymore. ::: ![](1476-7120-2-17-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Successful BVP documented by Tissue Tracking in apical two chamber view. A) Before BVP, there is a dilated ventricle with reduced systolic displacement (max displacement = 8 mm) in the septum and DLC in the inferior wall (no colour-coding) B) After three months of BVP, there is a reduction of left ventricular dilatation (reverse remodelling), increased tracking values, a basal-apical gradient and no DLC regions anymore. ::: ![](1476-7120-2-17-9) ::: Patient selection for BVP ========================= Only limited data are published concerning prospective echocardiography based patient selection for BVP. Bordachar et al. \[[@B29]\] performed a prospective study to identify TDI parameters that would predict the benefit of upgrading right ventricular pacing to BVP. 26 patients with normal LVEF and RVP and 16 patients with CHF and RVP were included. EMD was defined as the interval between the stimulation spike and the onset of the S wave. An intra-ventricular EMD of \>50 ms identifies patients with significant asynchrony. No correlation between asynchrony and QRS width was seen in the heart failure patients. ECG criteria would have misclassified 44% of the patients for mechanical ventricular asynchrony. This study has defined relevant asynchrony but did not assess the hemodynamic or electromechanical effects after upgrade to BVP nor effects on morbidity and mortality. Retrospective analysis after BVP ================================ Several studies were performed to retrospectively correlate markers of asynchrony to benefit from BVP. Lafitte \[[@B30]\] has included 15 patients with idiopathic DCM and a QRS of more than 140 ms (NYHA III-IV, LVEF \< 35%, LVEDD \> 60 mm) for BVP. Measurement of EMD was performed at baseline and after one month. This study has found that BVP reduces EMD in the lateral left ventricular wall. In another study \[[@B25]\], 49 patients with heart failure (QRS \> 130 ms, LVEF \< 35%, NYHA II-IV) were included. Retrospectively, intra- and interventricular and the combined index of asynchrony (=the sum of left and right ventricular asynchrony) were assessed at baseline and after 6 months of BVP by pulsed wave TDI. The cut-off-values for LV-asynchrony was 60 ms (56 ms for RV-LV-asynchrony and 102 ms for the \"sum-asynchrony\"). By definition, patients with a relative increase in LVEF of more than 25% were classified as responders to BVP. Receiver-operating characteristics (ROC) analysis showed that the degree of echocardiographic asynchrony is superior to QRS width in predicting hemodynamic and clinical improvement after BVP compared to QRS duration or conventional echo data. In 82% of the patients, the benefit of BVP could have been predicted echocardiographically. The role of TDI and 3D echo on the long term (1 year) outcome after BVP was evaluated in 25 patients \[[@B19]\]. The extent of DLC in the basal segments at baseline predicted the long-term efficacy of BVP. The LV base DLC was reduced from 18,7% to 8.1% after BVP. In concordance with other studies, the QRS duration failed to predict BVP efficacy \[[@B28]\]. The myocardial segments with the best resynchronization after BVP were studied in 18 patients with an LVEF \<35% and a QRS width of \>120 ms (NYHA III-IV). Color tissue Doppler velocity imaging was performed from the apical four chamber view at baseline and one month of follow-up after BVP \[[@B31]\]. Peak velocities and regional time differences in basal and mid septal segments were compared to the corresponding lateral segments. At baseline, a regional asynchrony of 42 ms in the basal sites (only 14 ms in the mid left ventricular site) was measured. After one month of BVP, a reduction of asynchrony was seen in only the basal segments but not in the mid segments. In conclusion, it was suggested that hemodynamic improvement is mainly in basal sites. Reverse remodeling and improved synchrony after 3 months of BVP was evaluated in 25 patients \[[@B32]\]. Asynchrony was assessed as time-to-peak regional sustained systolic contraction (=Ts). After three months, a homogenous left ventricular delay of Ts, improved interventricular synchrony and a reduced isovolumic contraction time and increased diastolic filling time were documented. These beneficial effects were reversible after withholding BVP. In a univariate analysis, systolic dyssynchrony was the only independent predictor of reverse remodeling after three months \[[@B33]\]. One recent study has compared the value of TDI and SRI and post-systolic shortening in the prediction of reverse remodeling after BVP: The previously introduced asynchrony index (=Ts-SD) based on Tissue Doppler velocity data has the highest predictive value of reverse remodeling after BVP. PSS has predictive power only in non-ischemic heart failure. In ischemic heart failure, PSS seems not to be a marker for reverse remodeling but rather reflects viability and is therefore not altered by BVP. SRI imaging techniques did not predict reverse remodeling after three months of BVP \[[@B33]\]. This is in contrast to previously published data \[[@B34]\]. Kanzaki has introduced the synchrony index, which is defined as the correlation coefficient of linear regression of velocity of septal and lateral mitral annular region. This index showed an increase after 6 months of BVP paralleled by increased LV contractility \[[@B35]\]. One study \[[@B36]\] has retrospectively evaluated the use of TSI to predict the acute response to BVP in 29 patients. The acute benefit to BVP was defined as a \>15% increase in echocardiographically measured stroke volume 48 h after device implantation. A difference of \>65 ms in time-to-peak velocity in anteroseptal and posterior segments in the apical long axis view was associated with acute improvement after BVP. However, the ability of TSI to predict long-term improvement after BVP needs further evaluation. Guidance for implantation ========================= TDI could play a role in identifying patients during catheterization procedures that will profit from BVP. Catheterization studies have shown that the beneficial effects of BVP begin almost immediately \[[@B37],[@B38]\]. But systematic evaluation with TDI-technique is currently ongoing. Furthermore, TDI can assist in finding the optimal pacing site for the coronary sinus lead. In 31 patients, it was documented that LV-stimulation on the site of longest EMD had the best benefit of BVP. The regional asynchrony was assessed by pw-TDI and the pacing site was determined fluoroscopically \[[@B39]\]. Lateral and postero-lateral LV lead positions were retrospectively found to improve left ventricular hemodynamics \[[@B40]\]. Optimal programming of biventricular device after implantation ============================================================== AV-time programming ------------------- An AV time is considered to be optimal when the end of the A wave coincides with the complete closure of the mitral valve \[[@B41]\]. An optimal AV time setting of the pacemaker can improve systolic function \[[@B42]\]. However, there is only limited published data assessing the optimal AV time in patients with BVP. Optimization of the interventricular delay ------------------------------------------ The optimal delay between the right ventricular and the coronary sinus stimulation is yet unknown. One study compared simultaneous versus sequential BVP in 29 patients. The optimum interventricular delay was found by maximum reduction of DLC as measured by Tissue Doppler and Tissue Tracking. An optimum sequential BVP could significantly reduce the extent of DLC compared to simultaneous pacing \[[@B43]\]. Patients with atrial fibrillation ================================= About one third of patients with heart failure have atrial fibrillation. The large trials, however, have only included patients in sinus rhythm. Only small studies with controversial results were performed in patients with atrial fibrillation and LBBB. Leclercq has performed one study in 59 NYHA III patients with chronic atrial fibrillation, a slow ventricular rate and the need for permanent pacing (VVI-paced QRS width of \>200 ms). Due to a high drop out rate, the results did not show a significant increase in 6-min-walk distance after BVP \[[@B44]\]. Larger trials are needed to evaluate BVP for patients with atrial fibrillation. Preliminary own results ======================= We have performed a double-blind cross-over study in our clinic to assess the use of new echocardiographic techniques in BVP. Patients (n = 40) with a QRS \>140 ms and a LVEF \<35% received an InSyncICD 7272 (Medtronic, Minneapolis, Minnesota, USA). Preliminary results (n = 8) after two years demonstrate a reduction of the septal-posterior delay from 264 (±23) msec to 234 (±34) msec (p \< 0,05) and a stabilization of clinical (NYHA class improvement) and hemodynamic status (EF and LV volumes). The study is ongoing. The following video loops underline the utility of TSI and Tissue Tracking to document improvement of synchronicity after BVP. In [additional file 1](#S1){ref-type="supplementary-material"} shows asynchrony before BVP implantation in apical four chamber view by TSI. In [additional file 2](#S2){ref-type="supplementary-material"} the effect of BVP is shown. in [additional file 4](#S4){ref-type="supplementary-material"} shows the acute changes of BVP as documented in this video loop by Tissue Tracking from apical four chamber view compared to baseline ([additional file 3](#S3){ref-type="supplementary-material"}). The long-term effect of BVP after six months is illustrated in [additional file 5](#S5){ref-type="supplementary-material"} (baseline) and [additional file 6](#S6){ref-type="supplementary-material"} (after 6 months). Conclusion and future perspective ================================= Many controlled and uncontrolled studies have demonstrated that new echocardiographic tools to determine myocardial asynchrony in heart failure patients will help to select patients for BVP help guidance of implantation and optimize device programming. To date, all studies employing tissue Doppler date were performed retrospectively. No prospective studies that have selected patients for BVP according to echocardiographic evaluation of asynchrony were performed yet. The ongoing CARE-HF study incorporates echocardiographic criteria of asynchrony in a subset of patients with a QRS of 120--150 ms \[[@B45]\]; results are not expected until 2005. The criteria of asynchrony in this study are (1) aortic pre-ejection delay \>140 ms, (2) the mechanical interventricular (pw aortic valve vs. pulmonary valve) delay \>40 ms and (3) the demonstration of left ventricular post-systolic contraction by M-mode and/or Tissue Doppler. Unresolved issues include different opinions regarding the various elements of asynchrony and their contribution to the pathophysiology and progression of heart failure. There is a lack of consensus about the best asynchrony marker for patient selection. There is evidence that ischemic and dilated cardiomyopathy might have different selection parameters for BVP. The practical consequences for patient selection and/or implantation site of the lead are currently under investigation. There are only limited echocardiographic data regarding the programming of the optimal interventricular (V-V) delay. There are no data concerning the long-term effect (i.e. years) of BVP on hemodynamics, amelioration of mitral regurgitation, reverse remodeling and mortality. Another area of uncertainty is the selection of patients for BVP without electrical (QRS \< 120 ms) but with mechanical asynchrony. Abbreviations ============= BVP Biventricular Pacing DCM Dilated Cardiomyopathy DLC Delayed longitudinal Contraction EMD Electromechanical Delay IVMD Interventricular Mechanical Delay LBBB Left Bundle Branch Block PSS Post-Systolic Shortening SRI Strain Rate Imaging TDI Tissue Doppler Imaging Ts Time-to-peak myocardial contraction TSI Tissue Synchronization Imaging Ts-SD Standard deviation of time-to-peak myocardial contraction Authors contributions ===================== F Knebel and AC Borges have performed the literature review and have prepared the manuscript. RK Reibis, have performed echocardiographic examinations for this article. HJ Bondke and J Witte and G Baumann have selected patients for BVP. HJ Bondke and J Witte have implanted the biventricular pacing devices. All authors have read and approved the final version of the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 TSI in a patient with LBBB before BVP: The lateral segments show a delayed time-to-peak velocity (red colour). Apical four chamber view. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 TSI post-implantation: There is only green colour coding indicating synchronous contraction of all segments from apical four chamber view. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Tissue Tracking without BVP : There are reduced displacement values and no basal-apical gradient in the septal segments and DLC in the lateral wall (no colour-coding) indicating asynchrony of the lateral wall. Apical four chamber view. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 Acute effect with BVP \"on\" (Tissue Tracking): There are increased displacement values, a basal-apical gradient. Apical four chamber view. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 Tissue Tracking before BVP : There are reduced displacement values and no basal-apical gradient in the septal segments and DLC in the lateral wall. Apical four chamber view. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 Long-term effect after 6 months of BVP: Reduction of left ventricular dilatation, increased displacement values, a basal-apical gradient. Apical four chamber view. ::: ::: {.caption} ###### Click here for file :::	PubMed Central	2024-06-05T03:55:48.351354	2004-9-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521694/", "journal": "Cardiovasc Ultrasound. 2004 Sep 15; 2:17", "authors": [ { "first": "Fabian", "last": "Knebel" }, { "first": "Rona Katharina", "last": "Reibis" }, { "first": "Hans-Jürgen", "last": "Bondke" }, { "first": "Joachim", "last": "Witte" }, { "first": "Torsten", "last": "Walde" }, { "first": "Stephan", "last": "Eddicks" }, { "first": "Gert", "last": "Baumann" }, { "first": "Adrian Constantin", "last": "Borges" } ] }
PMC521695	Background ========== In both plants and animals, during early meiosis in normal cells there is a clustering of all or most of the telomeres of the entire chromosome set to a single region on the nuclear membrane \[[@B1]-[@B3]\]. This meiotic looping of chromosomes with clustered ends has been termed the bouquet arrangementwhich appears synchronously with synapsis of bivalents. The reason why the telomeres attach to the nuclear membrane in meiosis is not dependent on the presence of normal numbers of TTAGGG repeats and, in fact, still occurs in late generation Terc-/- mice without detectable pantelomere repeats \[[@B5]\]. In plants the meiotic telomere clustering can be inhibited by colchicine \[[@B6]\] but a polarization still remains within the nuclei such that microtubules and nuclear pores are still arranged in a region that normally would face the telomere cluster on the opposite side of the nuclear membrane \[[@B3]\]. In mammals the bouquet arrangement seen at early meiosis occurs with some minor differences between males and females \[[@B4]\] but has disappeared in both by diplotene/dictyotene. For mitosis there is much less data on the position or possible associations of telomeres and subtelomeres. However, the spatial arrangement of chromosomes at mitotic interphase has been studied intensively \[[@B7],[@B8]\] but there are few studies with data on the principles that dictate nuclear organization. Nagele et al.\[[@B9],[@B10]\] using whole chromosome paints on fixed normal diploid human cells described a radial array (rosette) of prometaphase chromosomes where the chromosomes were apparently arranged in two tandemly linked haploid sets. That interphase chromatin formed ring-like shapes was already known \[[@B11],[@B12]\] but Nagele et al.\[[@B9],[@B10]\] proposed that there was a chromosome order in each of the haploid sets in diploid cells during mitosis which was thought to be reversed with respect to one-another. A chromosome order was also described as being present at interphase in non-cycling cells \[[@B13]\] where the nuclear organization seems to be fundamentally different from that in dividing cells \[[@B14],[@B15]\]. From observations in triploid cells, Nagele et al.\[[@B10]\] proposed that the three haploid sets were spatially arranged, two with the chromosomes arranged in tandem and the third with a reversed chromosome order. The relationship of subtelomeric regions to these concepts of a chromosome order within a radial chromosome array is less clear. Stout et al. \[[@B16]\] studied subtelomeric chromosome regions at interphase and showed that compared to interstitial chromosome sites, subtelomeres showed an increased number of somatic pairings. By FISH within living cells, Molenaar et al.\[[@B17]\] were able to demonstrate that these telomeric associations are dynamic. The rate of telomeric associations apparently depends on the stage of the cell cycle. Nagele et al.\[[@B18]\] utilising a telomere-specific peptide nucleic acid probe has demonstrated that the prevalence of such telomeric associations is far higher at interphase in non-cycling cells than in their cycling counterparts. In the present study we examine the telomere associations in mitotic interphase in human non-cycling cells of diploid or triploid karyotype. The cell types used were from skin, fetal cartilage, and long-term culture of chorionic villi but colchicine and hyptotonic treatments were avoided during cell harvest because of the potential effect of disrupting any topology present \[[@B6],[@B19]\]. We report a new finding, the detection of looped chromosomes in mitotic G1 by the intrachromosomal tethering of short-arm (p) and long-arm (q) telomeres. This new finding has implications for the understanding of the normal dynamics of chromosome behaviour at interphase but also for the processes involved in telomere capture. Results ======= Fluorescence in situ hybridization (FISH) performed with the various subtelomere probes (Table [1](#T1){ref-type="table"}) gave discrete signals in all experiments attempted. Figure [1](#F1){ref-type="fig"} shows FISH of cells probed with the p-subtelomeres labeled red and q-subtelomeres labeled green for chromosomes 4, 5, 7, 10, 17, and 20, arranged respectively in figure [1A,1B,1C,1D,1E,1F](#F1){ref-type="fig"} (diploid cell line CG04-0743BBRS) and for chromosomes 18, 12 and 6 in figure [1G,1H,1I](#F1){ref-type="fig"} (triploid cell line CG01-2042YA). The proportion of p-q associated signals is shown in Table [2](#T2){ref-type="table"}. The frequency of p-q subtelomere tethering ranged from 76--85% in the diploid cells but was a little less in the triploid cells (58--94%). Not all signal pairs were tethered. The percentage of diploid cells with allp-q signals tethered was 46--72% as compared to 33--58% in the triploid cells. This would be expected with more opportunity for interhomologous tethering in the triploid nuclei with an extra chromosome. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Origin and derivation of the telomere clones used in the study. ::: Clone Chrom Supplier ------------- ----------- -------------- 1186B18 3p Flint 196F04 3q Incyte 36P21 4p Incyte 963K6 4q Flint 189N21 5p Incyte 240G13 5q Incyte 62I11 6p Incyte 57H24 6q Incyte 164D18 7p Incyte 3K23 7q Incyte 43N06 9p Incyte 112N13 9q Incyte 306F07 10p Incyte 137E24 10q Incyte 496A11 12p Flint 221K18 12q Incyte 2111b1 17p ATCC 362K4 17q Flint 52M11 18p ATCC 964M9 18q Flint 1061L1 20p Flint 81F12 20q Incyte Note:The host strain was E. coli DH10B with kanamycin resistance in all cases except the clone 2111b1 (17p probe) which was ampicillin resistant. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Rate of tethering in non-cycling cells at G1 interphase of p (short arm) to q (long arm) subtelomeric signals in single homologues. ::: Chromosome Genome of cells Numbers and \[Percentage\] of p-q signal pairs\\ tethered (95% confidence limits) Numbers and \[Percentage\] of cells with [all]{.underline} p-q signals tethered\# (95% confidence limits). ------------ ----------------- ------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------ 4 Diploid 50/60 \[83\] (71--91%) 10/14 \[72\] (42--92%) 5 Diploid 123/148 \[83\] (76--89%) 26/41 \[63\] (47--78%) 7 Diploid 86/113 \[76\] (67--84%) 26/45 \[58\] (42--72%) 9 Diploid ND\* ND\* 10 Diploid 92/108 \[85\] (77--91%) 24/35 \[69\] (51--83%) 17 Diploid 62/76 \[82\] (71--90%) 13/28 \[46\] (28--66%) 20 Diploid 72/90 \[80\] (70--88%) 14/24 \[58\] (37--78%) 3 Triploid 25/30 \[83\] (65--94%) 5/10 \[50\] (19--81%) 6 Triploid 49/61 \[80\] (68--89%) 12/22 \[55\] (32--76%) 12 Triploid 57/82 \[70\] (58--79%) 8/24 \[33\] (16--55%) 18 Triploid 65/85 \[77\] (66--85%) 12/28 \[43\] (25--63%) ND-not determined. \* Cross-hybridization between 9q and 18p prevented analysis. \\ p-q subtelomere signal pairs scored, irrespective of cell numbers. \#Separate scoring of discrete cells only; i.e. diploid cells with clearcut tethering of both p (short-arm) to q (long-arm) subtelomere signal pairs and triploid cells with all three p-q subtelomere signal pairs tethered. ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A-IIntrachromosomal tethering of the subtelomeres of each single homologue in diploid and triploid non-cycling interphase nuclei at G1. FISH of diploid (A-F) or triploid interphase nuclei (G-I) from the following cell lines: CG04-0743BBRS (diploid) derived from skin and CG01-2042YA (triploid) derived from CVS. These non-cycling cells were probed with p-subtelomeric probe (labelled with spectrum orange) and q-subtelomeric probe (spectrum green) for (A) chromosome 4; (B) chromosome 5; (C) chromosome 7, (D) chromosome 10, (E) chromosome 17, (F) chromosome 20, (G) chromosome 18, (H) chromosome 12, (I) chromosome 6. The proportion of p-q tethered signals is shown in Table 2. In each case the majority of cells (76%--85%) showed pairing of short-arm and long-arm subtelomeres from single homologues often arranged on opposite sides of the interphase nucleus. Note also that an interphase topology is exhibited such that oval rosettes of chromatin can be seen in the present study in figs 1H, 1I, an elongated rosette in fig 1C, and off-centre rosettes in figs 1A, 1B. ::: ![](1475-9268-3-3-1) ::: The triploid cells were used to test the likelihood that intrachromosomal pairing of subtelomeric signals was occurring rather than the pairing of p and q signals with the q and p signals of the other homologue(s). As can be seen in figure [1G,1H,1I](#F1){ref-type="fig"}, there were three p-q tethered signal pairs in the triploid interphase nuclei. Figure [2](#F2){ref-type="fig"} shows single arm subtelomeric probes from three different chromosomes demonstrating that there is no linkage of positioning (chromosome order) between non-homologues. This is evidence challenging the claims of haploid groups being present at the interphase of non-cycling cells. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### A-FChromosome homologues at G1 in nuclei of non-cycling cells are spatially arranged without respect to non-homologues. Same cell line and same cell harvest as the triploid cells probed in Fig 1. Two combinations of three subtelomeric probes (see Table 1 for clones) are shown hybridized to triploid cells. In fig 2A-C the nuclei are probed with three single subtelomere probes from 4p (spectrum orange); 18q (spectrum green) and 6p (both spectrum orange and spectrum green labels, i.e. yellow signal). In figs 2D-F, the nuclei are probed with three subtelomere probes labelled 5p (spectrum orange), 12q (spectrum green), and 20p (spectrum orange and spectrum green, i.e. yellow signal). Note: There was no segregation into haploids sets of chromosomes at G1 interphase. Homologues were regularly arranged without any defined relationship to non-homologous signal groups; i.e. haploid sets of interphase chromosomes distributed to separate nuclear regions do not appear to exist. Note also there is a low frequency of isolated non-homologous associations: between 4p and 6p (Fig 2A); between 6p and 18q (Fig 2C), and between 5p and 20p (Fig 2D). ::: ![](1475-9268-3-3-2) ::: Further confirmation that the p-q tetherings in figure [1](#F1){ref-type="fig"} were from single chromosomes is shown in figure [3](#F3){ref-type="fig"}. Figure [3A,3B,3C](#F3){ref-type="fig"}, shows chromosome 15 interstitial loci (diploid cell line CG04-0743BBRS) probed together with the 15 alpha centromeric probe, and a 15q subtelomeric locus. Separate chromosome domains surround the subtelomeric signals (Fig [3A,3B,3C](#F3){ref-type="fig"}). Similarly for chromosome 12 (Fig [3D,3E,3F](#F3){ref-type="fig"}), using the same diploid cells (CG04-0743BBRS), subtelomeric probe pairs are defined to occur within the two separate chromosome domains by jointly using WCP12. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### A-FLooped chromosomes in G1 arrested cells: the distribution of tethered subtelomeric signals coincides with chromosome domains. Diploid non-cycling cells harvested after confluence arrest. The diploid cells are from the same cell line as in Fig 1 (i.e. CG04-0743BBRS). Fig 3A-3C shows diploid cells probed for chromosome 15 with CEN15 (larger signal spectrum green); SNRPN at 15q12 (spectrum orange); PML at 15q22 (spectrum orange); and subtelomeric 15q probe (smaller signal spectrum green). Note: The two chromosome 15 domains coincide with and envelop the 15q subtelomeric signals (there is no currently recognized specific 15p subtelomeric sequence and hence no 15p subtelomeric probe). Fig 3D-3F shows diploid cells probed for chromosome 12 with the subtelomeric probes for 12p (labeled with spectrum orange) and 12q (labeled with spectrum green) and the WCP chromosome 12 (the spectrum green smear). Note: The three chromosome 12 domains as defined by the (directly labeled) WCP12 envelop the three tethered subtelomeric probe pairs. This confirms that the telomeric tethering represents looped chromosomes. ::: ![](1475-9268-3-3-3) ::: Discussion ========== Evidence for short-arm and long-arm subtelomeres of the one homologue associating --------------------------------------------------------------------------------- This study shows that the pairings of red/green signals from the subtelomeres of the short-arm and long-arm respectively occur at high frequency in these non-cycling diploid nuclei. In many cases the association is so close that the subtelomeric signals are superimposed (e.g. figure [1A,1F](#F1){ref-type="fig"}). The pairs of red/green, p/q signals are from a single chromosome with the two diploid homologues arranged on different sides of the nucleus. This has been shown in this study in several ways. Firstly, it is highly likely that separate looped chromosomes are involved since the paired subtelomeric signals occur with small chromosomes (chromosome 17, 18, 20), intermediate chromosomes (7, 9, 10, 12) or large chromosomes (3, 4, 5) and are observed in triploid as well as diploid cells. Indeed the wide separation of the two subtelomeric signals from pairs of homologues (e.g. fig [1B,1D](#F1){ref-type="fig"}) supports the present interpretation that the telomeric tetherings of p-q signal pairs are intrachromosomal and not between homologues. Secondly, when interstitially located probes are used, for example on chromosome 15 (Fig [3A,3B,3C](#F3){ref-type="fig"}) in diploid cells, two distinct chromosome domains are seen. Thirdly, when subtelomeric probe pairs are used with a WCP probe for example on chromosome 12 (Fig [3D,3E,3F](#F3){ref-type="fig"}) in diploid cells, two distinct chromosome domains are seen that envelop the two tethered pairs of subtelomeric regions. In diploid nuclei the pairs of tethered subtelomeric signals are distributed to two areas and in triploid nuclei (fig. [1G,1H,1I](#F1){ref-type="fig"}), the tethered signals are distributed to three areas. If the signal pairings were between the short-arm from one homologue with the long-arm of another it is especially unlikely in the triploid cells that the chromosomes could span the diameter of the nucleus consistently. This is especially unlikely in light of the finding by Nagele et al.\[[@B10]\] that the nucleus normally exhibits a rosette of (chromosome rich) chromatin with a less dense central core (doughnut shape). If inter-homologous telomeric associations were the explanation for the regular p-q signal pairings then, especially in triploid cells, chromosomal threads would have to be arranged in very complex formations across the chromatin poor cores of rosettes. Finally, there is separate evidence that there are small non-overlapping chromosome territories at interphase in mammalian cells \[[@B20],[@B21]\] where the chromosomes are extended but not entwined. In the present study we have also been able to show the presence of these interphase chromosome domains both with the use of several probes spanning the length of chromosomes (e.g. Fig [3A,3B,3C](#F3){ref-type="fig"}) or with chromosome paints (e.g. Fig [3D,3E,3F](#F3){ref-type="fig"}). Nagele et al.\[[@B18]\] showed that there were very few coincident telomeric associations (TA\'s) in rapidly cycling mitotic cells. However, these authors showed \[[@B18]\] that in non-cycling cells there was a high rate of double associations, and a lesser frequency of triple and quadruple associations or unassociated telomeres. These authors \[[@B18]\] concluded that the replicative status of the cells was the prime determinant in the level of telomere associations. The finding of a high intrachromosomal p-q telomere association rate in the present study probably explains the underlying high telomere association rate described by Nagele et al.\[[@B18]\]. In that study \[[@B18]\], a universal telomere probe was used so the specificity of the associations, if present, was unrecognisable. In the present study, there was a high (\~80% but not saturated) rate of intrachromosomal pairing with only \~20% of telomeres unpaired with their homologous subtelomere. These two studies can be reconciled if the apparently (\~20%) unpaired subtelomeres (present study) were actually associated with non-homologous subtelomeres. Fig [2](#F2){ref-type="fig"} shows the presence of an underlying low rate of non-homologous telomere tetherings in these G1 arrested cells. Regulation of telomere associations ----------------------------------- In early meiotic cells the presence of the normal numbers of universal TTAGGG sequences is not required for massed telomere clustering \[[@B5]\]. A complementary finding was reported by Nagele et al. \[[@B18]\] who showed that in late passage mitotic cells the number of telomere associations (TA\'s) did not increase during progression to late passage crisis. This indicates that telomere shortening did not increase the number of TA\'s. Since the pantelomeric repeats occur at [all]{.underline} telomeres, the specific intrachromosomal association presently observed also cannot be due to their presence. Neither can the mechanism of tethering be related to chromosome specific subtelomeric sequences since the two homologues with identical sequences remain separated (Fig [1](#F1){ref-type="fig"}). There clearly are similarities between the looping of chromosomes seen in the present non-cycling mitotic cells and in the chromosome bouquets of early meiosis \[[@B2],[@B3]\]. These two apparently disparate phenomena may be related. If the synapsis of bivalents, unnecessary in mitotic cells, was removed from the meiotic bouquet arrangement mechanism, the intrachromosomal tethering of separated homologues as presently observed is what may be left. This mitotic looping may have been originally present since meiosis is believed to have evolved from mitosis. Chromosome topology at interphase --------------------------------- The global organisation of the interphase nucleus has provoked the interests of cell biologists for several decades but detecting the presence of any macromolecular domains has been challenging \[[@B8]\]. Nagele et al.\[[@B9],[@B10]\] was able to confirm with Feulgen staining and FISH that the chromosomes were arranged in rosettes, a ring of chromatin with partly-condensed chromosomes, which persisted through mitosis and was even maintained in the daughter cells at telophase. Oval rosettes can be seen in the present study in figs [1H,1H](#F1){ref-type="fig"}, and [3D](#F3){ref-type="fig"}; an elongated rosette in fig [1C](#F1){ref-type="fig"}, and off-centre rosettes in figs [1A,1B](#F1){ref-type="fig"}, and [2B](#F2){ref-type="fig"}. Through the use of FISH with chromosome specific alphoid probes and whole chromosome paints, Nagele et al.\[[@B10],[@B13]\] attempted to show that chromosomes in the rosettes appeared to be in an orderly arrangement in both diploid and triploid cells. These authors interpreted this order as specifically positioned haploid sets \[[@B9],[@B10],[@B13]\]. The pairing of subtelomere signals in non-cycling cells at interphase, as in the present study, is in some aspects consistent with these prior observations though we do not accept that haploid sets are spatially segregated and we found no evidence for an interphase chromosome order in the non-cycling cells of our cell lines. We have repeated this work with centromeric probes (not shown) and again there was no evidence of haploid groups or of a regular chromosome order though widely spaced homologous centromeric signals are usually observed (with respect to each chromosome considered separately). With respect to telomeric tethering in cycling cells (at G2) no such p-q telomeric tethering pattern is present in our observations of lymphocytes (not shown) and the only associations are of sister chromatids. That most lymphocytes are at G2 can be observed by the doubled signals representing sister chromatids (not shown) which is in contrast to the single (chromatid) signals in the unreplicated G1 nuclei (see fig [1](#F1){ref-type="fig"}). With centromeric and painting probes, Nagele et al.\[[@B9],[@B10],[@B13]\] detected the presence of what they believed to be haploid sets of chromosomes in both diploid and triploid cells with the sets on opposite sides of the nucleus. In some cell shapes (e.g. elongated, polymorphic, or lenticular shaped cells) this regular order was obscured but in spherical nuclei it was mostly evident. Whereas there is often a spatial separation of the telomeric signals from the various homologues of the diploid or triploid G1-arrested cells in the present data (see fig [1](#F1){ref-type="fig"}) there was no evidence for a chromosome order or haploid groups in the cell nuclei (fig [2](#F2){ref-type="fig"}). In the explanation of Nagele et al.\[[@B13]\] the haploid sets these authors proposed represented maternal and paternal chromosome contributions. In the present data each set of identical homologues (two in diploid or three in triploid cells) appear to be arranged without respect to those of other chromosomes (fig [2](#F2){ref-type="fig"}), i.e. the spatial arrangement is not an interchromosomal phenomenon. This means the theoretical haploid sets of chromosomes described by Nagele et al.\[[@B9],[@B10],[@B13]\] probably do not exist. Figure [2](#F2){ref-type="fig"} illustrates the two experiments performed in the current study to address the possible existence of haploid sets. These comprised examining the chromosome order for the single telomeres 4p (labelled with spectrum orange -- Vysis, Downers Grove, Illinois), 18q (spectrum green), and 6p (spectrum green and spectrum orange, i.e. yellow signal) jointly hybridised to the same confluence arrested cells, and in a second experiment: 5p (spectrum orange label), 12q (spectrum green), and 20p (spectrum green and spectrum orange) hybridised to a second slide of triploid cell nuclei. These cells are from the same harvest as those shown to display the interphase topology of p-q intrachromosomal subtelomere tethering. In these latter results, homologous subtelomeres were regularly arranged without any defined relationship to non-homologous signal groups. This demonstrates that there is no interchromosomal order transferable between nuclei and challenges the concept of the presence of haploid sets within these non-cycling cells. In the view of Nagele et al.\[[@B10]\] the dual odd topology that he observed with (i) homologues arranged on opposite sides of the nuclei (diploid cells) or regularly arranged around the nucleus (triploid cells), and (ii) a chromosome order possibly manifesting as \"haploid sets\" may just be a relic of fertilization. Whereas, in our view, these intrachromosomal telomeric tetherings with the resulting looped chromatids must be important to normal cell function. Possible relationship of telomere tetherings to telomere captures ----------------------------------------------------------------- As reviewed by Ballif et al.\[[@B23]\] there are two general pathways whereby chromosomes can acquire a new telomere and thus become stabilised. Firstly, by \"telomere healing\", i.e. the direct addition of telomeric repeats by: (i) telomerase \[[@B24]\] or by (ii) telomerase-independent recombination-based mechanisms \[reviewed in \[[@B25]\]\]. The second pathway is by \"telomere capture\" in which a chromosome acquires a telomere from another chromosome or chromosome end \[reviewed in \[[@B23]\]\]. Telomere captures are observed in two forms, those that are within the one homologue or intrachromosomal telomeric captures or transpositions \[[@B22],[@B23]\], and those between non-homologues \[[@B26]\]. Ballif et al.\[[@B23]\] considered four different models for telomeric captures involving the p and q arms of a single homologue (intrachromosomal captures). These telomeric captures where the telomere from one chromosome arm is deleted and replaced by a telomere from the other arm of the homologous chromosome were termed intrachromosomal transpositions of telomeres\[[@B22]\] because of the uncertainty that simple reciprocal translocation was involved in this type of telomere capture. Ballif et al. \[[@B23]\] suggested that breakage induced replication(BIR), reviewed in Kolodner et al.\[[@B28]\], was the most likely model for these intrachromosomal captures based on their observation that there was observed heterozygosity between the duplicated ends on the one chromosome. This mechanism was initially described by Reddel et al.\[[@B27]\] under the unwieldy name \"alternative lengthening of telomeres mechanism\". Ballif et al.\[[@B23]\] suggested that BIR simply copied the sequence from the other end of the same homologue. Furthermore, that obligatory crossing-over during meiosis would mean that heterozygosity between duplicated ends would occur in many cases. The detection in the present study for the first time that in non-cycling mitotic cells in G1 most short-arm and long-arm telomeres from the one chromosome are tethered together is a likely staging point for mitotic events as per the fourth model of telomere capture reviewed in Ballif et al.\[[@B23]\]. This fourth model is that of the present authors in a prior study \[[@B22]\]. In the explanation of that fourth model by Ballif et al. \[[@B23]\], telomere capture was believed to occur by a pre-meiotic interhomologous exchange. The imbalanced chromosome was then generated through normal meiotic recombination. This (model) thus resulted firstly in a balanced translocation, termed telomere transpositionby Daniel et al.\[[@B22]\] since reciprocaltranslocation was unproven. This translocation relocated the telomeres to the opposite chromosome arm and then by recombination the result was a duplication of one telomere on both chromosome ends and a deletion of the other. For this model to be correct a high frequency of balanced telomeric translocations would have to occur. These would be observed as large pericentric inversions and are rarely reported -- see review in Daniel, 1988 \[[@B30]\]. However, the transposition of telomeres to opposite chromosome ends resulting in large pericentric inversions would not be easily noticed during FISH in many cases. This is in contrast to translocations between non-homologues which are very obvious to an observer in a FISH study. In this connection, for telomere translocations between non-homologous the rate of clinically ascertained balanced translocations has been reported as very high. Flint and Knight \[[@B26]\] record that for the several types of (non-homologous) telomeric rearrangements: unbalanced translocations account for 54% of cases; deletions for 39%; and duplications for 6%. According to Flint and Knight \[[@B26]\] in almost all cases unbalanced translocations occur because a parent carries the balanced form. When the data used to draw this conclusion are scrutinised, see De Vries et al.\[[@B29]\] it includes many rearrangements that are microscopically detectable, i.e. essentially regular reciprocal translocations. Such latter rearrangements are not really \"telomere captures\", are often familial, and would be expected to be associated with a high rate of balanced carriers. In our experience (Greg Peters and Luke St Heaps -- CHW Telomere Study Group) we have not detected a balanced carrier of a telomere capture of either the intrachromosomal type or the interchromosomal type. In our view the issue of the frequency of balanced telomere rarrangements needs to be revisited since telomere captures are technically sub-microscopic telomere rearrangements. This data impinges on the likelihood that BIR is the preferred method of telomere capture \[see that view expressed in ref \[[@B23]\]\]. Since with the BIR model immediate recombinants are formed with no balanced carriers, if balanced (telomere capture) carriers are frequently reported, then BIR is ruled out as the common mechanism of telomere capture. This judgement currently cannot be performed without a more rigorous approach to the whole data set and additional assessment of the de novo or alternative origin of telomere rearrangements. Conclusions =========== A topology for telomeres was detected where looped chromosomes were present at G1 interphase in confluence arrested cells. It was shown that the p and q telomeres of each chromosome in G1 cells associate frequently (80% pairing) in an intrachromosomal manner which was confirmed by studying chromosome domains with interstitial probes (chromosome arms) and WCP probes. It was found that homologues were regularly arranged without any defined relationship to non-homologous signal groups; i.e. there was no apparent chromosome order on different sides of the nuclei and no segregation into haploid chromosome sets was detected. The normal function of this high frequency of intrachromosomal telomeric pairings is unknown but a potential role is likely in the genesis of telomere captures whether of the intrachromosomal type or between non-homologues. In our view, these intrachromosomal telomeric tetherings with the resulting looped chromosomes arranged in a regular topology must be important to normal cell function since non-cycling cells in G1 are far from quiescent, are in fact metabolically active, and these cells represent the majority status since only a small proportion of cells are normally dividing. Materials and methods ===================== Cell culture ------------ Cell lines were retrieved from liquid nitrogen, washed in Dulbecco\'s phosphate buffer (DPB) and reconstituted in Hams F10 medium. The following lines were used: a diploid skin fibroblast line CG04-0743BBRS with karyotype 46,XX derived from fetal cartilage and a triploid 69,XXX cell line CG01-2042YA of diandric origin derived from mesodermal cells of a chorionic villus biopsy. These were cultured until they reached confluence via contact inhibition. At this stage the cells exhibit a number of swirls of closely packed cells in parallel. They were then severally prepared for trypsin harvest usually 48 hours after the last media change without colcemid/colchicine treatment and without the usual 0.075 M KCl hypotonic treatment. The cells were trypsinised off and fixed three times in 3:1 methanol to glacial acetic acid and were stored at room temperature (R.T.) in fixative for 1--3 days. This period allowed some mild acidic digestion of the chromatin and spreading of the nuclei when slides were prepared. At the end of the storage period, cells were rewashed once with fresh fixative and dropped onto glass slides as per routine techniques, and stored on trays in a low humidity cabinet until used for FISH. Choice of probe and probe label ------------------------------- The identity of the probes used in the study is shown in Table [1](#T1){ref-type="table"}. The clones containing the DNA for the subtelomere probes were obtained from three sources: Incyte Genomics (Fremont, California); Dr Jonathan Flint (John Radcliffe Hospital, Oxford, U.K.), via Dr David Mowat, or the ATCC, (Manassas, Virginia). All were grown in Luria Broth (LB) with kanamycin by standard techniques unless specified otherwise (Table [1](#T1){ref-type="table"}). Plasmid DNA was extracted with QIAGEN midi kits as per the manufacturer\'s instructions except that DNA elution was achieved at 60°C overnight. Probes were all labelled by nick translation (using VYSIS kit and the fluorophores spectrum orange and spectrum green, Vysis, Downers Grove, Illinois) as per the manufacturer\'s instructions. Fluorescence in situ hybridization (FISH) ----------------------------------------- Slides were pretreated with a combined Pepsin/Rnase step. This was performed by prewarming RNAse and pepsin to 37°C, 200 μl of RNAse (0.1 mg in saline/sodium citrate -- 2xSSC) was dispensed onto each slide, coverslipped and incubated at 37°C for 40 minutes in a humidified chamber. Coverslips were removed and slides washed twice for 5 minutes in 2xSSC at room temperature (RT). Slides were briefly drained and 200 μl of pepsin (0.2% in 0.01 M HCl) was placed on the slides, coverslipped and incubated at 37°C for 30 seconds. Coverslips were removed and slides were washed twice for five minutes in phosphate buffered saline (PBS) at RT. Fixation was with 6% paraformaldehyde in PBS, by dispensing 200 μl/slide, and adding a coverslip for 2 minutes at RT. Slides were then washed twice for five minutes in PBS at RT, dehydrated through 70, 90 and 100% ethanol for 3 minutes/wash at RT, and air dried. Probes in hybridisation mix were stored at -20°C, removed and thawed for 30 minutes; dispensed onto slides, covered with 15 mm diameter coverslips, and sealed with liquid rubber -- art cement. Joint denaturation was achieved at 75°C for 5 minutes on a Omnigene hot plate, transferred to a humidified hybridization chamber at 37°C and hybridised overnight. After this the coverslips were removed. Post-hybridization washes were 0.4 SSC/0.3% NP40 at 73°C for 2 minutes then quickly transferred to 2xSSC/0.1% NP40 at RT for 1 minute. Slides were counterstained in DAPI and then rinsed and air dried. When ready, slides were mounted in antifade (2.3% DABCO in 40% glycerol/0.02 M TRIS-HCl) and covered until fluorescence examination. Slides were examined on a Zeiss Axioscop 20 fitted with a Zeiss fluoarc light source and images captured on an Applied Imaging Cytovision station using the false colours that are attributed by the software. Scoring of signal pairings to detect telomere tethering ------------------------------------------------------- Initially, the subtelomere probes were labelled in Spectrum Orange for all short arms and Spectrum Green for all chromosome long arms. Cells were separately probed with the two subtelomeric probes for a single chromosome at the one time. Probe pairs were used for the subtelomeres of chromosomes 1, 3, 4, 5, 6, 7, 9, 10, 12, 17, 18, and 20. Images were captured for a large number of cell groups for each chromosome and pairings were scored on the captured images. Signals were interpreted as paired if the distance between signals was 10% or less of the greatest diameter of the nucleus (many cells were oval in shape). In addition to the above subtelomeric probe pairs, other probes were used to investigate the frequency of non-homologous tetherings (subtelomeric probes 1p and 9q, see Table [3](#T3){ref-type="table"}) and the coincidence of the subtelomere tetherings and interphase chromosome domains (Fig [3](#F3){ref-type="fig"}). For the latter experiment the following Vysis probes were used: PML (promyelocytic leukemia locus) mapping to 15q22, SNRPN (small nuclear ribosomal protein locus) mapping to 15q12 -- both labelled with spectrum orange; CEN15 (a probe for alpha centromeric sequences specific to chromosome 15) labelled with spectrum green. In addition, the chromosome 12 subtelomeric probes, i.e. 12p (labelled with spectrum orange) and 12q (spectrum green), and the WCP (Vysis whole chromosome painting probe) for chromosome 12 (spectrum green). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Rate of subtelomeric tetherings of non-homologues in G1 of non-cycling cells. ::: Pairs of telomeres tested for tethering No (%) of signal pairs tethered 95% confidence limits ----------------------------------------- --------------------------------- ----------------------- 1p telomere; 9q telomere\* 7/109 (6.4) 2.6--12.8% Note: This is a control for chromosome specific p-q subtelomere signal pair tethering (Fig 2, Table 2). For this experiment the 1p subtelomere was labelled with spectrum orange and the 9q subtelomere with spectrum green. Subtelomere tethering between these non-homologues was not increased over chance expectation. The telomere pair\* 1p and 9q were chosen because they represent one of the most frequent telomere translocations reported (Lisa Shaffer, personal communication 2004). Up to 10% of signal associations in a two colour matrix can be regarded as random (VYSIS guidelines for interphase FISH scoring). There was no departure from randomness for tethering with respect to these two pairs of non-homologous subtelomeres. ::: Additional experiments to detect a regular chromosome order reflecting the possible existence of haploid sets regularly arranged around the nuclei -------------------------------------------------------------------------------------------------------------------------------------------------- Two such experiments were performed in the current study (see fig [2](#F2){ref-type="fig"}). These comprised examining the chromosome order for the single subtelomeres (see Table [1](#T1){ref-type="table"} for clones) 4p (labelled in spectrum orange), 18q (spectrum green), and 6p (spectrum green and spectrum orange, i.e. yellow signal) jointly hybridised to the same triploid cells. In a second experiment subtelomeres were labelled as follows: 5p (spectrum orange), 12q (spectrum green), and 20p (spectrum green and spectrum orange, i.e. yellow signal) hybridised to a second slide of diploid/triploid cell nuclei. Authors\' contributions ======================= AD designed the study, captured and analysed all FISH signals, and drafted the manuscript. LH performed all growing of probes, labelling of probes, and most probe hybridizations. Acknowledgements ================ This work was supported by NCH Grant SG8682. We thank Dr Greg Peters of the Cytogenetics Department, Children\'s Hospital Westmead, and Dr Roger Reddel of the Children\'s Medical Research Foundation for helpful discussion. We also thank Dr Zhanhe Wu and Jill Cross of the Cytogenetics Department, Children\'s Hospital at Westmead for the culturing and provision of archived cell lines and the FISH probe hybridisations for chromosome 15, and Robert Tamas of the CHW-IT Department for assistance with collating the figures.	PubMed Central	2024-06-05T03:55:48.354083	2004-9-29	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521695/", "journal": "Cell Chromosome. 2004 Sep 29; 3:3", "authors": [ { "first": "Art", "last": "Daniel" }, { "first": "Luke", "last": "St Heaps" } ] }
PMC521696	Background ========== Oxythioquinox (Morestan™ or OTQ, Bayer Corp) is a prototypical pesticide that was first used in 1968 on crops such as apples, pears, cucumbers, and gherkins. However, its use was later confined to non-food crops, limiting exposure to nursery and greenhouse employees. OTQ is a member of the quinoxaline class of pesticides, which also includes chlorquinox and thioquinox. Principal agricultural use of OTQ was limited to the states of California, Washington, Florida, New York, Pennsylvania, Ohio and Michigan \[[@B1]\]. The use of OTQ in the United States was voluntarily cancelled in 1999 with stocks in use until 2001, although OTQ is still listed in many different regions for use as an insecticide \[[@B2],[@B3]\]. Further, OTQ is still in use today in other areas of the world, including Australia and the Caribbean \[[@B4],[@B5]\]. Scientifically acceptable toxicity studies of OTQ are sparse. Early in vivostudies in rats found alterations of a variety of metabolic enzymes following OTQ exposures, including alkaline phosphatase \[[@B6]\]. Although not directly analyzed, the results of this study combined with an earlier study \[[@B7]\] suggest a direct effect of OTQ on succinate dehydrogenase, or other enzymes with a thiol group. Further work by Carlson et al (1970) \[[@B7]\] showed that, although OTQ has a low acute toxicity, cumulative exposure to this pesticide is not well-tolerated by exposed animals, with the majority of damage found in the liver of these animals. Further studies looking mainly at hepatic enzyme function found that OTQ exposure inhibits some hepatic enzyme functions \[[@B8]\]. OTQ was shown to be a carcinogen and hepatotoxin in laboratory animals in later studies \[[@B9]\]. OTQ has also been classified as a probable human carcinogen \[[@B10]\]. However, potentially carcinogenic exposures have already occurred and OTQ is still in use outside of the US, its mechanism of action remains of interest. In addition, OTQ has been shown to have an inhibitory effect on cytochome P450s, enzymes known to have a pivotal role in carcinogen metabolism \[[@B8],[@B11]-[@B13]\]. Although early studies focused on hepatic effects, carcinogenic potential may occur in other tissues as well. While there have been multiple pesticides implicated in breast cancer, no studies have been published related to OTQ exposure and human cancer incidence \[[@B13]\]. This study is similar to work currently being carried out to determine gene expression profiles of a variety of environmental agents, including chemicals, physical agents and physiologic stresses \[[@B14]-[@B19]\]. The National Institute of Environmental Health Systems (NIEHS) has recently funded a consortium to expand the study of gene expression profile signatures for various chemicals, with the long-term goal to use these analyses in a validated gene expression profile signature database. A recent report by Shan etal. 2002 \[[@B20]\] is a good example of the differences between gene expression profile signatures for two chemicals in an animal system. This report profiled gene expression in rat carcinomas induced by two carcinogens, 2-amino-1-methyl-6-phenylimidazo \[4,5-b\]pyridine (PhIP) and 7,12-dimethylbenz \[a\]anthacene (DMBA), and was able to show that while both chemicals altered expression in some genes in a similar fashion, each induced unique gene expression patterns. The primary goal of this study was to look for consistent changes common to all donors that could potentially be used as biomarkers of exposure, creating a gene expression profile following OTQ exposure in normal human cells. Given previous research in hepatoxicity, ideally normal human liver cells would have been used as a model system. However, we needed a normal human tissue that was readily accessible. Therefore, gene expression was studied in primary normal human mammary epithelial cells from four different donors in response to OTQ exposure. Current microarray analysis of pesticide exposure focuses on animal studies, not allowing for analysis of inter-individual variation. With the use of microarrays including clinical diagnosis, genetic susceptibility to disease and treatment, the need to determine the potential role of inter-individual variation on gene expression patterns in all potentially exposed tissues. Genes found to be altered in multiple cell strains could be considered biomarkers for individual populations. Given the small number of cell strains used, follow-up analysis in a larger number of samples is required to confirm any potential biomarkers. Biomarkers derived from this study could potentially be used in future epidemiology studies analyzing the effects of pesticide exposure. Further, this gene expression profile could be compared to those of other pesticides and of known carcinogens to develop a more detailed mechanistic definition of these chemicals. Methods ======= Cell culture ------------ Primary normal human mammary epithelial cells (NHMECs) were derived from tissues salvaged at reduction mammoplasty obtained though the Cooperative Human Tissue Network (National Cancer Institute and National Disease Research Interchange). Development and characterization of cell strains was achieved using standard methods \[[@B21]\]. Cells were grown in MEGM media (Clonetics, Cambrex, Pittsburgh, PA) at 37°C and 5% CO~2~. OTQ treatment ------------- Treatment was performed on cells in passage six at 70% confluency, as routinely performed in our laboratory. Preliminary studies analyzed a range of OTQ concentrations (0 -- 12.5 μM) selected based on previous research \[[@B6],[@B7]\] and time points (0 -- 24 hours), and showed maximum effect of OTQ on p53expression with minimal toxicity at 2 hours with a final concentration of 6.25 μM. Cells were treated by diluting the stock OTQ/DMSO mixture in media and adding this solution to aspirated cells, allowing even exposure to all cells. DMSO (0.001%) alone was used as a vehicle control. At the end of the treatment period, cells were removed for RNA isolation. Cell viability was determined by Trypan Blue exclusion assay. Indirect immunofluorescence --------------------------- Confluent cells in passage five were trypsinized and plated on eight-well slides (LabTek II Slide System, Nunc, Naperville, IL). These cells were grown to 70% confluency before being treated with OTQ (6.25 μM) for 15, 60 and 120 min or DMSO (0.001%) for 120 min. At the end of treatment, the media was aspirated and the cells were fixed with methanol. Slides were then stained with anti-p53antibody (1:1000, DO-1, Santa Cruz Biotech, Santa Cruz, CA) and incubated overnight at 4°C. The next day, the media was again removed and the secondary antibody (1:1500, goat anti-mouse FITC, Santa Cruz Biotech) was incubated for one hour at room temperature. Slides were washed in triplicate with phosphate buffered saline (PBS) and cover slips were added. Slides were dried for one hour at room temperature before viewing, using the laser scanning confocal microscope BX50 (Olympus), and quantitative analysis was performed \[[@B22]\]. Relative p53expression was quantified between different time points and strains by determination of the area under the integrated intensity curve (Fluoview, Olympus, B & B Microscopes, Pittsburgh, PA). Microarray analysis ------------------- Microarray analysis was performed in duplicate using the HuGeneFL high-density oligonucleotide microarrays (Affymetrix™, Santa Clara, CA). Protocols from the Affymetrix Expression Analysis Technical Manual were followed RNA was isolated from cells with Trizol (Gibco, Grand Island, NY), followed by purification with RNEasy Mini Kit (Qiagen, Valencia, CA). Spectrophotometer measurements were required to give a 260/280 ratio of 1.9--2.1 for use in microarray analysis. Double-stranded cDNA was then synthesized from total RNA (Superscript Choice System, Invitrogen, Carlsbad, CA). An in vitrotranscription (IVT) reaction (Enzo, Farmingdale, NY) was then performed to produce biotin-labelled cRNA from the cDNA. Excess biotinylated dUTPs were removed by RNEasy Mini Kit before being fragmented and added to a hybridization cocktail, including Eukaryotic Hybridization controls (Affymetrix), BSA and herring sperm DNA (Gibco, Grand Island, NY) and biotinylated anti-streptavidin antibody (Vector Laboratories, Burlingame, CA). Hybridization on microarrays was performed for 16 hours at 45°C in the Gene Chip Hybridization Oven with rocker (Affymetrix). Microarrays were washed and stained using the protocol, as described in the Affymetrix Manual, with the GeneChip Fluidics Station 400 (Affymetrix). Arrays were then scanned with the Affymetrix Scanner (Hewlett Packard, Palo Alto, CA). Expression profiles were analyzed using Microarray Suite 5.0, MicroDB 3.0 and Data Mining Tool 3.0 (Affymetrix). Affymetrix arrays are produced using multiple 25-mer oligonucleotides (11--20 per target gene). Each oligonucleotide is created to match the selected region of the target gene (perfect match, PM), while a similar oligonucleotide is created altered in the 13^th^position to control for non-specific binding (mismatch, MM). Results are given in signal intensities with a p-value determined from perfect match/mismatch (PM/MM) intensities by Tukey\'s Biweight analysis. Each array was normalized to a scaling factor of 1500 to correct for array variation. All arrays for each cell strain were analyzed on the same day to minimize variation. Signal log ratio was determined by comparison of the signal intensities for the baseline (vehicle control) and the treatment array. This is computed using a one-step Tukey\'s Biweight method by taking a mean of the log ratios of probe pair intensities across the two arrays. This method helps to filter out differences due to different probe binding coefficients that may lead to false positives and/or negatives. A signal log ratio of zero represents no change in gene expression as a result of OTQ exposure. A signal log ratio of one is equivalent to a fold change of two between the treatment and control. The results described here are the average of both duplicates, with the average percent variability between duplicate arrays being 1.5% (the average difference found between duplicates, related to array to array variability as well as technical variability in processing the array). Only relative changes equal to or greater than 0.6 signal log ratio (SLR) were considered a significant change as a result of exposure. The biological significance of each change is determined with Wilcoxon\'s signed rank test with the Affymetrix software. Gene chip analysis was performed by self-organizing map (SOM) clustering, focusing on genes with a detection pvalue of 0.05 or less at one or more time points. Analysis was performed to comply with MIAME standards. Real-time polymerase chain reaction analysis (RT-PCR) ----------------------------------------------------- cDNA synthesized from each sample as in the Affymetrix analysis (Invitrogen) was used in a one-step RT-PCR analysis reaction. Analysis was performed in duplicate on the ABI 7700 cycler, with the SYBR Green Master Mix (ABI) and samples were normalized using both 18S and GAPDH expression levels for each sample. Primers were designed using Primer Express^®^(ABI) to yield unique fragments for each gene under study. Reactions were set up following recommended protocols using 100 pmol of each primer (Sigma-Genosys) and approximately 60 ng template per reaction. Reactions were performed in duplicate for each sample for 40 cycles (95°C/15 sec denaturing step; 60°C/1 min annealing/extension step). Fold change was determined based on average cycle threshold (C~T~) values for all duplicates and converted to signal log ratio. Results ======= Trypan blue exclusion test -------------------------- Trypan Blue was used to analyze toxicity by measuring cell viability for each cell strain for each treatment. The results showed a range of viability from 92--97% at all time points, except for the last time point in strain 3, which had a viability of only 65% at 120 min (results not shown). This decrease in viability at 120 min was not found to be directly correlated to p53 or p21 protein expression, or to any particular gene expression pattern. The dose of OTQ used was based on these findings. Indirect immunofluorescence --------------------------- Baseline p53protein levels were visually compared to those after treatment in each cell strain. Integrated fluorescence intensities were measured on each optical slice of cells. The fluorescence was determined as the area under the curve in arbitrary units (AU) (Figure [1](#F1){ref-type="fig"}). This result was compared between time points for each cell strain. An increase was seen in p53expression in all cell strains with increasing duration of an OTQ exposure. However, one strain (3) showed a 10-fold lower p53expression at each time point. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Quantitative analysis of immunofluorescence microscopy.Confocal microscopy analysis of p53expression. Integrated intensity measures were obtained from Fluoview and graphed with GraphPad Prism (GraphPad Software, San Diego, CA) to determine area under the curve as a measure of comparative p53protein expression for each treatment time point. Data is shown for each time point per cell strain on the ordinate and the intensity on the abscissa (arbitrary units, AU). There was an increase in p53protein expression for all of the strains tested in direct correlation with OTQ exposure. In one intermediate strain (3), p53expression was 10-fold lower at all time points. ::: ![](1476-069X-3-9-1) ::: DNA microarray -------------- DNA microarray analysis found no change in p53, despite the increase in p53protein levels observed by immunofluorescence. Studies of benzo \[a\]pyrene exposure in our laboratory have also found similar results for p53expression \[[@B23]\]. The effect of OTQ exposure on other cell cycle genes, however, was determined by DNA microarray analysis. Although inter-individual variation between donors in response to OTQ was evident, there were also some genes found to be increased consistently in all strains by microarray analysis (Table [1](#T1){ref-type="table"}). Self-organizing map (SOM) clustering was used to group genes with similar patterns of alteration in each of the strains. SOM analysis was performed following filtering of the total genes on the array, limiting the SOM analysis to only genes found to be present on at least one array analyzed. Following suggested analysis with the Affymetrix system, the default settings of the Affymetrix software were selected, including selecting threshold filtering (min = 20, max = 20000), row variation filtering (max/min = 3 and max-min = 100), and row normalization (mean = 0, variance = 1). Working with a 3 × 3 analysis to obtain 9 clusters generally gave an optimal amount of different clusters with less than 100 genes per cluster. From the SOM clustering analysis, genes altered by a signal log ratio of ± 0.6 or greater were chosen for closer study. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Genes altered following oxythioquinox exposure. Table represents data mined from HuGeneFL microarrays (Affymetrix). All genes selected have a signal log ratio of ± 0.6 unless otherwise noted. Representative genes for each group were selected based on their function and are shown here. ::: GenBank ID Name Peak Expression Level (SLR) Functional Class --------------------------------------------------------------- --------------------------------------------------------- ----------------------------- ----------------------- Genes increased in three or more strains (n = 13): U22028 CYP2A13 1.5 xenobiotic metabolism U20734 junB 3.42 transcription V01512 cfos 2.04 transcription S85655 prohibitin 0.75 cell proliferation S82240 RhoE GTPase 2.81 signal transduction M69043 MAD-3 mRNA encoding IkB-like activity 0.83 apoptosis M63573 cyclophilin 1.68 immune response U05861 Dihydrodiol dehydrogenase 2.52 xenobiotic metabolism Genes decreased in three or more strains (n = 23): L05624 MAP Kinase Kinase -0.67 signal transduction U18018 E1A enhancer -1.34 transcription X56681 junD -1.03 transcription X68836 S-adenosylmethionine synthetase -2.57 cell metabolism J04973 Cytochrome bc-1 -3.12 cell metabolism Genes altered in at least two of four cell strains (n = 189): X03484 raf oncogene 1.46 carcinogenesis M60974 growth arrest and DNA-damage-inducible protein (gadd45) 1.51 DNA damage M57731 Human gro-beta 1.85 immune response X66899 EWS 1.2 carcinogenesis Z29087\* Cyclin D1 Promoter 1.03 cell cycle control L10910 Splicing Factor CC1.3 0.62 RNA processing M83667 NF-IL6 Permeability Factor 1.55 transcription M27281 Vascular Permeability Factor -1.09 cell proliferation L28010 HnRNP F protein -0.55 RNA processing U72649 BTF2 -1.86 carcinogenesis M19267 tropomyosin -1.16 cardiac M38258 retinoic acid receptor gamma 1 -0.94 cell metabolism U42031 Immunophilin -1.41 immune response U67122 ubiquitin-related SUMO-1 -1.03 protein metabolism X70340 Transforming growth factor alpha -0.57 cell proliferation M34458 Lamin B -1.33 cell proliferation \No accession number was used by Affymetrix. This accession number most closely matches the probe description and sequence. ::: SOM clustering was used to show patterns of expression in each cell strain, and followed by further subclustering of those clusters of interest across all cell strains (Figure [2](#F2){ref-type="fig"}). This analysis was performed with Data Mining Tool 3.0 (Affymetrix). For comparison to earlier versions of the Affymetrix software, a signal log ratio of one is equal to a fold change of two. SOM clustering led to the selection of genes found to be altered, with the overall total of 215 genes (13 increased in 3 or more strains; 23 decreased in 3 or more strains). To be included, the genes must have a signal log ratio of ± 0.6, which is approximately a 1.5 fold change. Signal log ratio was used to linearize the data for ease of analysis. The full list of genes altered can be found at <http://www.cdc.gov/niosh/ext-supp-mat/weston/AlteredGenes.htm>. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### SOM clustering.Self-organizing map clustering groups\' genes by similar expression patterns. Each data point represents an OTQ treatment time point. Data points are plotted in order of cell strain with four points per strain (DMSO, 15 min, 60 min, and 120 min). Levels of expression are not given, as patterns of expression are based on relative expression levels. Panels 4 and 7 of this graph show genes increased at 15 minutes for one strain (3) follow similar pattern in second strain (4). Similar pattern is also seen in panels 6 and 8 of this graph. ::: ![](1476-069X-3-9-2) ::: From the full list of genes that fit these criteria, selected genes were chosen due to their potential role in carcinogenesis, whether by cell cycle control, immune response or other specific functions. Functions were described as annotated by NetAffx \[[@B24]\]. These genes are listed in Table [1](#T1){ref-type="table"}. Some genes were selected based on their possible role in disease as a result of pesticide exposure. Table [1](#T1){ref-type="table"} contains 8 genes increased in at least three of the four cell strains analyzed by 1.5 fold, as compared to the vehicle control and a list of 5 genes decreased in three or more of the four strains analyzed. Included in this list are two genes involved in the metabolic activation of endogeneous chemicals, cytochrome P4502A13 (CYP2A13)and dihydrodiol dehydrogenase. Although expression was slightly increased at one time point, the temporal patterns were slightly different (Figures [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"}). This increase in all cell strains suggests a potential role for these genes in OTQ metabolism. These results are confirmed at these time points as well as at 12 h and 24 h by RT-PCR (Table [2](#T2){ref-type="table"}). Genes also found to be increased in at least three of four cell strains analyzed include genes involved in transcription (junB, cfos), immune response (cyclophilin), and apoptosis (MAD-3). These genes showed a consistent increase in expression following exposure to OTQ (Table [1](#T1){ref-type="table"}). Genes that showed a consistent decrease in expression following exposure include signalling pathway genes (MAP kinase kinase), cell metabolism genes (S-adenosylmethionine synthetase, cytochrome bc-1), and transcription factors (E1A enhancer). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Expression pattern for CYP2A13.DNA microarray analysis of NHMEC strains. Analysis was performed as described on HuGeneFL high-density oligonucleotide microarrays (Affymetrix). Results are plotted as duration of exposure vs signal log ratio (SLR). Signal log ratio is a measure of comparative expression of the treatment vs. vehicle control (0.001% DMSO). Signal log ratio of one is equal to a fold change of two. A SLR of 0.6 (Fold Change \~1.5) was the arbitrary limit of our analysis. All genes given an Absent call by analysis software are shown with a SLR of zero. Asterisks indicate a statistically significant variation in expression from the control level as measured by Tukey\'s Biweight analysis. ::: ![](1476-069X-3-9-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Expression pattern for dihydrodiol dehydrogenase.DNA Microarray analysis of NHMEC strains. Analysis was performed as described on HuGeneFL high-density oligonucleotide microarrays (Affymetrix). Results are plotted as duration of exposure vs signal log ratio (SLR). Signal log ratio is a measure of comparative expression of the treatment vs. vehicle control (0.001% DMSO). Signal log ratio of one is equal to a fold change of two. A SLR of 0.6 (Fold Change \~1.5) was the arbitrary limit of our analysis. All genes given an Absent call by analysis software are shown with a SLR of zero. Asterisks indicate a statistically significant variation in expression from the control level as measured by Tukey\'s Biweight analysis. ::: ![](1476-069X-3-9-4) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### RT-PCR results. Table represents real-time PCR data from selected genes of interest. Time points used were 15 min, 120 min, 12 h and 24 h where shown. Results shown are results of replicate analysis with duplicate samples. Samples not analyzed represented by N/A. ::: Strain Gene SLR 15 min SLR 120 min SLR 12 h SLR 24 h -------- ------------ ------------ ------------- ---------- ---------- 1 BTF2 1.41 N/A 1.53 0.3 2 BTF2 -1.56 N/A -1.32 -1.74 3 BTF2 0.18 N/A -0.89 -16.61 4 BTF2 0.62 N/A -0.36 0.65 1 CYP2A13 5.58 6.01 3.84 10.6 2 CYP2A13 2.7 0.62 1.1 -0.6 3 CYP2A13 5.76 4.77 -0.6 -1.64 4 CYP2A13 5.45 1.1 0.33 -1.51 1 DDH 1.23 -0.67 1.75 1.47 2 DDH -0.47 -0.62 0.96 5.41 3 DDH 0.86 -3.47 2.01 -3.84 4 DDH 3.44 N/A 5.53 6.09 1 EWS -1.18 -1.84 0.69 -5.64 2 EWS -6.64 -0.4 1.32 1.54 3 EWS -0.42 -0.22 -1.79 -6.64 4 EWS N/A 2.92 0.78 0.86 1 GADD45 -0.09 -0.22 -1.32 -5.06 2 GADD45 -1.03 -1.69 -1.94 -1.32 3 GADD45 0.82 -2.32 -1.56 -4.32 4 GADD45 1.34 9.05 6.4 7.72 1 MAD-3 -0.71 N/A -0.15 -3.64 2 MAD-3 -1.89 N/A -1.06 -0.38 3 MAD-3 0.34 N/A -0.58 -6.64 4 MAD-3 4.4 N/A -4.38 0.51 1 PROHIBITIN 0.31 N/A -0.3 0.96 2 PROHIBITIN -0.6 N/A 0.3 -1.18 3 PROHIBITIN -1.06 N/A -1.94 -2.84 4 PROHIBITIN 2.3 N/A -1.09 0.92 1 RAF 0.01 -2.25 1.21 -5.64 2 RAF -1.51 -0.27 -1.43 -0.06 3 RAF 0.37 -3.84 -1.12 -7.67 4 RAF 1.33 7.57 4.8 7.22 1 RhoE 0.59 N/A -0.62 -0.71 2 RhoE 0.26 N/A -1.18 1.08 3 RhoE 1.55 N/A -0.94 -5.64 4 RhoE 2.35 N/A 1.06 2.05 ::: Some genes of interest were variably altered by strain, examples of which are also listed in Table [1](#T1){ref-type="table"}. These include genes involved in carcinogenesis (raf oncogene, GADD45, EWS, Cyclin D1) as well as immune response (immunophilin, gro-β), cell proliferation (TGFβ), and RNA processing (HnRNP F protein). Real-time PCR ------------- Real-time PCR was used to confirm and extend results seen by microarray analysis for selected genes. Following the original microarray analysis, patterns of some genes appeared to be changing at the latest time point (120 min), so extended time points were selected (12 and 24 h) to see a more complete expression profile for these genes. Due to limited amount of cDNA, genes were analyzed at 15 min and/or 120 min, and then analyzed at 12 and 24 h by RT-PCR. Extended time points were selected to look at specific genes found altered at the earlier time points. These genes were selected due to their function and/or pattern of expression, and determining their expression pattern at later time points was of interest. Results are shown in Table [2](#T2){ref-type="table"}. In the majority of samples, RT-PCR confirmed data found by microarray analysis for the genes listed. Some discrepancies are also shown, however, in these cases it is believed that the primer sequence is more specific to the gene in question than the probe sequence used on the array. However, extended time points, in some cases, showed that the results of early time points did not always continue to extended exposures. The RT-PCR results for prohibitin, DDH, and CYP2A13at 24 h showed a decrease in expression in some of the cell strains analyzed. Samples not available for RT-PCR analysis are listed as N/A in Table [2](#T2){ref-type="table"}. Discussion ========== The purpose of this study was to determine if microarray analysis of four normal human mammary cell strains with a known haplotype could be used to find biomarkers related to either exposure in general or the specific haplotype in question. The use of only four cell strains was determined to have enough power to provide basic information to lead to further study if necessary. This study would be followed up for specific genes of interest in a larger number of cell strains, preferably bypassing the more expensive and time-consuming microarray analysis for RT-PCR only. DNA microarray analysis was used to profile the cellular response to OTQ, a quinoxaline pesticide. Analysis revealed genes with common response across the four human cell strains studied as well as inter-individual variation in response. Using only four normal human cell strains, our goal was to discover any distinctly altered genes in response to OTQ exposure. Future analysis with a larger number of cell strains will be used to follow-up this analysis on specific genes of interest. The majority of studies looking at gene expression profiles have used animal models, limiting any knowledge obtained to genetically similar organisms. Analysis with normal human cell strains, like those described here, will give more information on inter-individual variation in response to various chemicals. This information will yield clues to the metabolic pathways of the specific chemicals, and this increased knowledge will aid in determining potential hazards in the environment and the workplace. Given the large number of pesticides in use today, further examination of the effect of these chemicals on individuals is warranted. Following exposure to OTQ, NHMECs showed alterations in genes involved in a variety of functions. These included xenobiotic metabolism, transcription, and DNA synthesis. Genes altered as a result of OTQ exposure in all strains analyzed included transcription factors like junBand cfos(the AP1 complex). A number of genes involved in carcinogenesis were found altered after exposure to OTQ, both induced and down-regulated. For example, prohibitinexpression is found to be up-regulated in most cell strains after OTQ exposure, with similar expression patterns associated with a decrease in cancer incidence \[[@B25]\]. Metabolism genes that were altered after OTQ exposure, like cytochome P4502A13 (CYP2A13)and dihydrodiol dehydrogenase, are involved in xenobiotic metabolism. It has been suggested that CYP2A13 is the main metabolic activator of 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK), a tobacco-specific nitrosamine \[[@B26]\]. The increase of CYP2A13begins to wane at the final time point tested, suggesting this alteration to be somewhat transient. This is an unusual expression pattern for a cytochrome p450, as genes in this family tend to show a gradual increase in induction, and an equally gradual decrease in expression. Further analysis by RT-PCR showed that this p450 was increased at later time points (12 h and 24 h, Table [2](#T2){ref-type="table"}). Another metabolic enzyme affected by OTQ exposure is dihydrodiol dehydrogenase. Dihydrodiol dehydrogenase is known to participate in activation of certain polycyclic aromatic hydrocarbons (PAHs), so its increase in the intermediate variant after OTQ exposure may result in an increase in PAH activation \[[@B27],[@B28]\]. Alterations of genes like this may suggest an indirect role for OTQ in carcinogenesis. Of these two genes, only CYP2A13 seems specific to exposure to OTQ. DDH has been found to be increased following exposure to various chemicals, including malathion, di-n-butyl phthalate, and benzo \[a\]pyrene (Gwinn in preparation, 2004) \[[@B23],[@B29]\]. Some of the early results at 15 min by microarray analysis may have been a consequence of the stress of exposure, regardless of the chemical. Extending analysis to later times determined whether results seen at these early time points were still valid after longer exposure to OTQ. RT-PCR results given in Table [2](#T2){ref-type="table"} show that in most cases, the extended time points showed a continued trend of expression (whether increased or decreased), except in some cell strains for DDH, prohibitin, and MAD-3. These results show a reverse of the early expression patterns at the later time points. Real-time PCR is a more specific method of analysis, as it only interrogates one gene at a time with primers designed uniquely to that gene. Conflicts in results between the two methods can generally be attributed to cross-reaction between probes designed for similar genes on the array. Sequence differences between the probes on the array and those used in RT-PCR may also play role in these results. The RT-PCR primers were selected specifically for the gene in question, while the probes on the array may not have been. Genes with sequence homology but with altered patterns of expression may not have been differentiated in the array analysis, but would be with the RT-PCR analysis. Inter-individual variation as a result of genetic polymorphisms in genes of interest would focus on specific at-risk worker populations. For example, the four cell strains analyzed in this study have been genotyped for a variety of genes, in particular those involved in cell cycle control and xenobiotic metabolism. Two of the four cell strains selected for analysis are heterozygote for the minor variant haplotype of p53, a cell cycle control gene (cell strains 3 and 4). Although no biological mechanism for the role of this variant in carcinogenesis has been defined, several studies associating this haplotype with various cancers support such role \[[@B30]-[@B35]\]. Analysis of genes altered in just those strains expressing this variant, including three genes involved in cell cycle control: raf oncogene(X03484), cyclin D1(Z29087), and BTF2(U72649) may further support an association between OTQ exposure, p53 variant status and carcinogenesis \[[@B36]-[@B38]\]. Of these, p53 has been reported to increase GADD45transcription in response to DNA damage, which is associated with an increase in cell cycle arrest and DNA damage repair, while increased levels of raf oncogenehave been associated with lung carcinogenesis \[[@B39]-[@B41]\]. Given the small number of cell strains used, this analysis needs to be extended to additional cell strains to determine the role of the p53variants in gene expression differences. Due to the expense of microarray analysis, this is performed in only a limited number of cell strains (4), and selected genes will be further analyzed with \\|RT-PCR in a larger number of cell strains. Over 80 cell strains have been established in our laboratory to date, with half of these having been genotyped for p53. However, given that this haplotype is found only in a limited portion of the population, this varied pattern of expression in key genes in cell cycle control may highlight a specific at-risk population. Searches for similar natural compounds to replace these potentially disruptive chemicals can also use gene expression profiles \[[@B42]\]. Profile comparisons to that of a natural pesticide may decrease the need for organophosphates. Comparison of OTQ\'s gene expression profile to that of well-defined chemicals, like benzo \[a\]pyrene, will yield important information about OTQ\'s role in both genotoxicity and potential carcinogenicity. A comparison between many expression profiles is needed to further define similarities and differences between this pesticide and known carcinogens and/or other pesticides. Conclusions =========== The overall goal of this project was to create a gene expression profile for OTQ or related pesticide analogues with the hopes of finding genes to be used as potential biomarkers of exposure. This expression profile may also be used to determine the final role of OTQ in carcinogenesis by comparing it to profiles of known carcinogens. It is possible that the main effect of OTQ exposure is not on the direct alterations in many genes, but on alterations in genes potentially involved in carcinogenesis, among them the examples of CYP2A13and dihydrodiol dehydrogenase*. The results shown here do not suggest a direct role of OTQ in carcinogenesis. They do, however, suggest OTQ exposure leads to an increase in expression of genes that do play a direct role in carcinogen; metabolism (for example, exposure to carcinogens like NNK and benzo \[a\]pyrene along with exposure to OTQ may lead to an increased incidence of tobacco-related cancer) \[[@B27],[@B43]\]. Discovery of genes altered following exposure to OTQ in human cell strains may aid in future epidemiology studies on pesticide exposures. Gene expression profiling can be used to yield genetic biomarkers of exposure that, after validation, could be used in a clinical setting for early determination of organophosphate exposure, increasing early treatment of pesticide illness and thereby increasing the recovery rate of exposed individuals. List of abbreviations used ========================== OTQ, oxythioquinox DDH, dihydrodiol dehydrogenase DMSO, dimethylsulfoxide SLR, signal log ratio DMT, Data Mining Tool™ MAS, Microarray Suite™ SOM, self-organizing map NHMEC, normal human mammary epithelial cell RT-PCR, real-time polymerase chain reaction Competing interests =================== None declared. Authors\' contributions ======================= MRG participated in the design of the study, and performed all experiments. DLW was responsible for the growth and maintenance of all cell strains used. AW conceived of the study and participated in the 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/content/backmatter/1476-069X-3-9-b1.pdf> Acknowledgements ================ We would like to thank Dr. Channa Keshava for assistance with microarray technology and Ms. Tanya Headley for implementation of the table dataset website. We would also like to thank Drs. Mark Toraason and David Weissman for their helpful comments on this manuscript, and Ms. Heather Marstiller and Ms. Glory Johnson for assistance with the manuscript format.	PubMed Central	2024-06-05T03:55:48.357638	2004-9-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521696/", "journal": "Environ Health. 2004 Sep 23; 3:9", "authors": [ { "first": "Maureen R", "last": "Gwinn" }, { "first": "Diana L", "last": "Whipkey" }, { "first": "Ainsley", "last": "Weston" } ] }
PMC521697	Background ========== Many patients do not eat and drink sufficiently during hospitalisation. Thus, 30--50% of the elderly patients are undernourished \[[@B1],[@B2]\] and most of these patients\' protein and energy requirements are not met \[[@B3],[@B4]\]. Their muscular tissue, including their heart and respiratory muscles, is adversely affected by this situation \[[@B5]\] and their immune function is suppressed \[[@B1],[@B6]\]. The clinical consequences include lassitude, difficulty in mobilising, prolonged convalescence \[[@B1],[@B7]\] and an increased risk of pressure wounds \[[@B8]\], phlebitis and infections \[[@B9],[@B10]\]. Patients often have reduced appetite, nausea or aversion towards certain types of food, which may partly explain the inadequacy of their food and liquid intake. Intervention studies have shown that by offering food or in-between meals rich in energy and protein, it is possible to increase the patient\'s protein and energy intake \[[@B11]-[@B14]\]. However, most of these intervention studies only use quantitative data. The present intervention study offers data, both quantitative on patients\' food intake and qualitative on the occupational groups\' attitudes and experiences in relation to the intervention, the food service and the nutritional care in general. These data can contribute to raise our knowledge of nutritional care in general and to identify issues crucial to an improvement of hospitalised patients\' food intake in particular. The first aim of this research was to examine to which extent standard hospital food met hospitalised medical patients\' protein and energy requirements. The second aim was to introduce nursing procedures focusing on the nutritional care based on the Danish nutritional recommendations for inpatients \[[@B15]\] to investigate the effect of this intervention on the patients\' intake of protein and energy. The third aim was to explore the involved occupational groups\' attitudes towards nutritional intervention and nutritional care in general. Particular attention was paid to the identification of problems possibly related to insufficient patient nutrition. Methods ======= Setting ------- The setting was an endocrinology ward with 49 beds and 3481 patients discharged during 2002 (divided into bed sections IA and IB) and a cardiology ward with 53 beds and 4542 patients discharged during 2002 (divided into bed sections IIA and IIB) \[[@B16]\]. All hospital food was produced in a central hospital kitchen and transported in heated containers to the bed sections where it was portioned out and served to the patients. Design of the study ------------------- Medical patients\' pre-intervention dietary protein and energy intakes were assessed by 72-hour weighed food records \[[@B17]\] at four bed sections (two wards) to include the appropriate number of patients. Before the intervention the bed sections at each ward was randomised to intervention or control. After a five-month intervention period, patients\' dietary protein and energy intakes were assessed to evaluate the effect of intervention. After intervention the occupational groups involved in the nutritional care and the food service at the two intervention sections were interviewed in focus groups or by individual interview. Participants ------------ Both acute and referred medical patients at all ages participated. The inclusion criteria were defined as: 1) the patient was not placed on a prescribed diet, 2) the patient had no contact with, or had not previously received dietary advice from a clinical dietician, 3) the patient did not belong to an ethnic minority, and 4) the patient was hospitalised for at least five days. Patients with dementia and patients who were severely mentally or physically impaired were excluded. Typical patient diagnoses included acute or chronic lung disease (e.g. chronic obstructive lung disease, asthma, bronchitis), acute or chronic cardiovascular disease (e.g. hypertension, angina, thrombosis, apoplexy), metabolic disorders (e.g. thyrotoxicosis, osteoporosis) or infectious disease (e.g. pneumonia, cystitis). The nursing staff selected the patients meeting all the criteria. The patients received oral and written information about the investigation underlining the voluntary nature of their participation. Three or four patients from a bed section participated at the same time, providing data for the food records. The occupational groups participating in the focus group interviews were: nurses, health care support staff and nurse aides on day or evening duty from one of the two intervention sections IB and IIB (four interviews), two nurses in charge from the two intervention sections (one interview), three maids from the two intervention sections (one interview), two clinical dieticians from the two intervention sections (one interview), and one catering officer from the kitchen (one individual interview). In total 26 informants participated in eight interviews. Patients\' characteristics -------------------------- Data on patient age, date of and diagnosis on hospitalisation, second diagnosis, oedema, dehydration, body weight on hospitalisation (if measured) were collected by the investigators from hospital records. Body temperature (if fever) was collected from the hospitals records during the 72-hour food recording. Patients\' body weights were recorded twice: at hospitalisation (or when they were included in the study) and on discharge. This weighing was standardised according to time of the day, the patients dress and the scales. The changes in body weight during hospitalisation were recorded for the patients not having oedema or dehydration. Patient height was measured and body mass index (BMI, kg/m^2^) calculated. The patient was asked about ability to chew and swallow and recorded as \'effortless\', \'slight difficulty\' or \'with difficulty\'. On discharge they ascribed to the meals during hospitalisation was recorded as \'very important\', \'of some importance\', \'almost no importance\' and \'no importance\'. Food records ------------ The patients had their food and drink weighed for 72 hours at breakfast, lunch, afternoon coffee and supper by the investigators. The patients, relatives or staff recorded the last-meal-of-the-day and individual between-meals as estimated records. The investigators contacted the patients three to five times a day to follow up on these estimated records. The weight of the between meals provided by visitors was estimated by weighing similar food items. The food items were weighed in the form received from the kitchen. Potatoes, mashed potatoes, sauce, meat, etc. were weighed separately. Standardized menus such as stew, open sandwiches, sandwiches, etc. were weighed in full. The total weight of the food items each patient was served was weighed before serving as was plate waste after the meal. Drinks were estimated and recorded when poured into a glass or feeding cup one centimetre below the rim. Dietary intake of protein and energy ------------------------------------ 24-hour food records were checked and coded to calculate protein (gram) and energy (kJ) intake by the investigators. The calculations were based on the data from the recipes used in the hospital kitchen and the Danish Database \'Dankost 2000\', which contains data from the Danish food tables \[[@B18]\]. Physical activity ----------------- Each patient\'s physical activity was recorded every hour during three days and nights (72 hours) in the period of food recording. It was recorded whether the patient was \'lying asleep\', \'lying awake\', \'sitting\', \'walking\' or \'training\'. For walking or training the approximate duration of activity was recorded as a fraction of an hour, and a factor of physical activity was estimated for each 24-hour period \[[@B15]\]. The investigators contacted the patients three to five times a day to follow up on the recording of the physical activity. Estimation of protein and energy requirements --------------------------------------------- Official Danish food recommendations for institutions propose that patients with chronic diseases have 1.0--1.5 gram of protein per kilogram body weight depending on the degree of stress metabolism \[[@B15]\]. A factor of 1.2 gram was used as an estimate of moderate metabolic stress \[[@B19],[@B20]\]. However, they did not allow for the underestimation of underweight and overestimation of overweight patients\' requirements. The calculations were adjusted accordingly in the following way: If the BMI was below 20, the recommended requirement was calculated as 1.5 gram per kilogram bodyweight per 24 hours. If the BMI was above 30, the recommended requirement was calculated as 1.0 gram per kilogram bodyweight per 24 hours \[[@B21]\]. The estimated energy requirement was calculated as \'basal metabolic rate\' (Harris Benedict equation \[[@B22]\]) x \'the factor of physical activity\' x \'the factor required to increase the body weight (if the BMI was below 20)\' or a \'factor of stress (if the BMI was above or equal to 20)\' \[[@B15],[@B21]\]. If the BMI was below 20, the factor 1.3 was used instead of the stress factor. The factor required to increase the body weight was an estimate of the amount of energy the patient was able to consume \[[@B15],[@B22]\]. The stress factor was applied in patients judged to have metabolic stress because of their pathological condition. The stress factor range was 1.1--1.4 for patients with chronic lung disease, chronic heart disease and apoplexy: severe infections were given a factor of 1.3. The stress factor was determined by the temperature and was set at 1.2 at a temperature of 38°C, 1.3 at 39°C and 1.4 at 40°C. Only one factor of stress was used, and the temperature stress factor had the highest priority \[[@B15]\]. The mean recorded protein (g) and energy (kJ) intake was compared with the estimated protein (g) and energy requirements (kJ), and the degree (in percent) to which the patient\'s 24-hour requirements were met. Intervention ------------ The nurses in charge from the two intervention bed sections IB and IIB received information a) specifying to which degree the patients\' protein and energy requirements were being met before intervention and b) detailing the Danish Recommendations for Hospitalised Patients \[[@B15]\]. In order to introduce and facilitate continuous staff monitoration of the patients\' nutritional status during their hospitalisation the following intervention procedures were formulated in collaboration with the two nurses in charge. Such monitoration would allow the staff to identify patients at risk of under nourishment and would secure continuous registration, which was seen as a precondition for optimising the patients\' uptake of nutrients. The procedures were formulated as one standard applying to be used to all non-diet patients admitted to bed sections IB and IIB: The patient\'s nutritional status is assessed on admission and during hospitalisation. As recommended by the nurses in charge three forms (A, B and C) with different purpose were made to support the staff in relation to the nutritional care. In Form Apatient data related to the nutritional care were collected upon admission: height, body weight, BMI, usual body weight and changes in body weight for a defined period (if possible), oedema or dehydration, the date and by whom the patient had been informed about the food service, the result of the first assessment of the nutritional status (result from form B), a short description of 1) the patient\'s ability to eat and drink, and 2) of the action taken by the staff 3) the date of the next assessment of the nutritional status. The form also allowed room for the results of the next five assessments of the patient\'s nutritional status. The purpose of Form Bwas to assess the patient\'s nutritional status/risk score and suggesting the action the staff could take. Actions were performed according to detailed English Standards \[[@B23]\]. The assessment parameters were body weight for height (BMI), appetite and ability to eat. The patient was assessed at \'low risk\' when BMI was normal, the appetite good and the patient fully independent. The patient was assessed at \'moderate risk\' when underweight but stable, the appetite poor, and the patient needed help with feeding or had some swallowing difficulties. The patient was assessed at \'high risk\' when severely under weight or actively lost weight, ate very little or have had no food for the last four meals, and was dependent on others for feeding or had severe swallowing difficulties. Form B allowed the patient\'s nutritional status to be recorded six times to ensure continuity of the assessment of nutritional status. A short guide to action was provided to the staff for each of the assessment categories; For the \'low risk\' patient \'no action necessary, but check weight weekly\'. For the patient at \'moderate risk\' the action could be \'check weight weekly, encourage with eating and drinking, replace missed meals with supplements and repeat score after one week and ask medical staff to refer patient to clinical dietician if no improvement\'. For the patient at \'high risk\' the action was to \'focus on encouraging with eating and drinking, replace missed meals with supplements and repeat score after three to four days and ask medical staff to refer patient to clinical dietician\'. In Form Cthe estimated record of the patient\'s protein and energy intake could be calculated and compared to data in the nutritional handbook describing what the food items contained of protein and energy. This handbook contained a standardised description of all meals delivered from the kitchen in household measurements (spoons, pieces, decilitre, etc.) and the estimated protein (g) and energy (kJ) content. Introducing the standard ------------------------ The investigators convened meetings with the nursing staff and the domestic helpers at the two intervention sections IB and IIB. The rationale of the standard was explained detailed and both oral and written instructions about the use of the forms were given. Four meetings were held in bed section IB and six in bed section IIB. At these meetings problems, ideas, etc. related to the standard and the forms were discussed and adjusted according to these discussions. The investigators contacted the staff in bed sections IB and IIB once or twice a week during the five-month investigation period to give support if wanted. The intervention at the two bed sections had no influence on the food production in the hospital. But before the intervention the kitchen produced a \'unrestricted diet\' to all patients not placed on a prescribed diet, which contain about 8250 kJ and 70--80 gram of protein with about 15, 41 and 43% of energy from protein, fat and carbohydrates \[[@B24]\]. During the intervention period the kitchen changed the production to two different diets to meet the Danish Nutritional Recommendations for Diseased People \[[@B15]\]; From the kitchen the diets were introduced in the following way; one diet for the elderly and people with little appetite -- the so-called \'hospital diet\' -- and one diet for all patients with ischemic heart disease and diabetes mellitus -- the so-called \'normal diet\'. The \'hospital diet\' contained about 10000 kJ and 90 gram of protein with 18, 40 and 42% of energy from protein, fat and carbohydrates. The \'normal diet\' contained about 9000 kJ and 80 gram of protein with 10--15, 30 and 55--60% of energy from protein, fat and carbohydrates \[[@B24]\]. The changes in the diets were introduced to the staff by the clinical dieticians. Besides these diets different commercial and no-commercial protein- and energy supplements, stewed fruit, soup etc. were available from the kitchen. Statistical methods ------------------- The number of patients required was calculated in the following way:The clinically relevant difference between the average extent to which the patient\'s protein and energy requirement was met before and after the intervention was estimated to 15% \[[@B25]\]. Patients\' dietary protein and energy intakes were estimated to lie 0--50% below their requirements (standard deviation (SD) 12.5--15.0%). A 5% significance level was chosen and the power was chosen to lie at 90%. The t-test was used to calculate an appropriate sample size for the control and intervention groups, viz. a minimum of 21 patients. The dietary protein and energy intake was calculated as a 24-hour mean (SD) for each patient and for each group (SD) of patients at each bed section. The outcome measure was the percentage degree to which the patient\'s actual protein and energy requirement was covered compared with his/her estimated requirement. Confidence intervals for the outcome measures were estimated. The effect of intervention for independent groups of patients were tested by one-way analysis of variance (ANOVA) using the SPSS version 9.0. The assumptions of independence, normality and identical variances were fulfilled. Analyses of covariance were described for non-comparable variables for the four patient groups after intervention. The interview in the occupational groups ---------------------------------------- An interview guide was designed for each of five occupational groups: 1) nurses, health care support staff and nurse aides (four interviews), 2) charge nurses, 3) maids, 4) clinical dieticians and 5) one catering officer from the kitchen \[[@B26]\]. In the interview the investigator focused on the informants\' actions, attitudes, experiences and reflections in relation to the intervention and nutritional care. Focus group interviews were considered the most appropriate form of data collection given the intent of the study \[[@B27]\]. All the 26 informants shared experience from the intervention study and from the situations where patients\' meals were served. Eight focus group interviews were carried out at the hospital during the working hours in rooms familiar to the voluntary informants. The focus group interviews were tape-recorded with the permission of the informants, who were informed that they could read the transcribed interview, should they wish so. The qualitative data were analysed as a text. Ethical approval ---------------- The study fulfilled the declaration of Helsinki II and was approved by the Local Scientific Ethics Committee. Results ======= Patient characteristics ----------------------- Food records were completed for 48 patients before and 60 patients after intervention. Table [1](#T1){ref-type="table"} summarises the baseline characteristics of the participating patients. The patient groups were comparable with regard to BMI, stress factor and ability to chew and swallow. The average age of the medical patients was 72 ± 11 years. Before the intervention 17 patients out of 22 lost body weight. After the intervention 20 patients out of 37 lost body weight (table [1](#T1){ref-type="table"}). Before the intervention, 56% of the patients participating in the study were weighed on admission (defined as within 48 hours from their arrival to the bed section). After intervention 52% of the patients in the control sections and 45% in the intervention sections were weighed on admission by the staff. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Summarised baseline characteristics of participating patients before and after intervention. Values are group averages (standard deviation (SD)) unless otherwise stated. ::: Medical ward I Medical ward II ---------------------------------------------------------------- -------------------------- ------------------------------- --------------------------- -------------------------------- Number of patients (women/men) 12 (5/7) 12 (10/2) 10 (7/3) 14 (12/2) Age, years (SD) 74 (13) 68 (14) 72 (7) 70 (10) Length of stay, 24 hours (SD) 25 (21) 33 (29) 24 (18) 25 (18) BMI, kg/m^2^(SD) 26.4 (4.2) 26.6 (5.1) 26.1 (6.2) 24.7 (6.0) BMI, women, kg/m^2^(SD), 27.5 (4.2) 27.0 (5.5) 26.5 (6.8) 25.3 (6.2) BMI, men, kg/m^2^(SD), 25.7 (4.4) 24.4 (1.5) 25.2 (5.7) 21.1 (2.5) Change of body weight per 24 hours, gram (SD) (n)\# 20 (100) (4) -78 (83) (7) -154 (107) (5) -6 (174) (6) Medical ward I Medical ward II After intervention Bed section IA Control Bed section IB Intervention Bed section IIA Control Bed section IIB Intervention Number of patients (women/men) 16 (9/7) 14 (10/4) 13 (9/4) 17 (7/10) Age, years (SD) 74 (12) 73 (13) 71 (9) 73 (9) Length of stay, 24 hours (SD) 26 (20) 24 (17) 14 (7) 16 (10) BMI, kg/m^2^(SD) 22.2 (6.3) 22.1 (3.7) 25.9 (7.1) 24.9 (4.9) BMI, women, kg/m^2^(SD) 21.8 (4.7) 22.1 (4.3) 26.6 (8.4) 21.9 (4.5) BMI, men, kg/m^2^(SD) 22.6 (8.0) 22.3 (2.2) 24.4 (3.6) 27.0 (4.0) Change of body weight per 24 hours, gram (SD) (n) \# -72 (184) (8) 11 (108) (11) -105 (140) (8) -87 (238) (10) Number of patients receiving \'Hospital diet\'/\'Normal diet\' 10/6 11/3 9/4 10/7 \# Patients who take diuretics and patients with dehydration or oedema are excluded. ::: Patient requirement and protein and energy intake ------------------------------------------------- In table [2](#T2){ref-type="table"} the average degree to which protein requirements were met before and after intervention are summarized. In table [3](#T3){ref-type="table"} the corresponding figures for energy requirements. There were no significant pre-intervention differences between the groups concerning the average degree to which their estimated protein (p = 0.918) and energy (p = 0.367) requirements were met. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Dietary intake of protein, estimated requirement of dietary protein and degree to which need for dietary protein per 24 hours was covered before and after intervention. Values are group averages (standard deviation (SD)) unless otherwise stated. ::: Medical ward I Medical ward II --------------------------------------------------------------- -------------------------- ------------------------------- --------------------------- -------------------------------- Before intervention Bed section IA Status Bed section IB Status Bed section IIA Status Bed section IIB Status Dietary intake of protein in grams per 24 hours (SD) 63 (26) 56 (10) 55 (20) 59 (23) Estimated need for dietary protein in grams per 24 hours (SD) 85 (13) 79 (12) 79 (18) 84 (13) Estimated need for dietary protein covered in per cent (SD) 73(27) 73(15) 71(24) 72(30) 95 % confidence interval 56--90 63--82 54--89 55--89 Medical ward I Medical ward II After intervention Bed section IA Control Bed section IB Intervention Bed section IIA Control Bed section IIB Intervention Dietary intake of protein in grams per 24 hours (SD) 44 (20) 61 (26) 49 (15) 49 (20) Estimated need for dietary protein in grams per 24 hours (SD) 74 (12) 72 (13) 81 (17) 85 (14) Estimated need for dietary protein covered in per cent (SD) 60(26) 85(31) 62(19) 56(19) 95 % confidence interval 46--73 67--102 51--74 46--66 ::: The results of the intervention was different at bed section IB and IIB; The intervention significantly improved the degree to which the energy and protein requirements were met among patients in intervention section IB compared with patients in the control sections IA and IIA (protein p = 0.009 and energy p = 0.010). On average, the former had an intake of 85% of their calculated protein requirement and 103% of their energy requirement. In intervention section IIB, the patients only had an intake reaching 56% of their protein and 76% of their energy requirement. These values were on average much lower than for patients in section IB and they were comparable to those obtained in the control sections IA and IIA. Analysis of co-variance for the non-comparable variables age, patient mobility, BMI, type of diet and number of bed-days showed no significant effect on the outcome measure for the degree of meeting the patients\' requirement of protein and energy. The patients ability to chew and swallow, and the importance of the meals to the patients during hospitalisation were comparable in the four groups of patients before and after the intervention. In the control sections the diet met 61% of the patients\' protein and 75% of their energy requirements after intervention. These levels were not significantly different from those recorded before the intervention, but 11% and 14% lower than before the kitchen changed the diets. The intervention and the occupational groups -------------------------------------------- During the intervention period, the nursing staff in bed section IB used the forms for assessing the nutritional care of three patients. In intervention section IIB the forms was used assessing the nutritional care of 17 patients. The patients nutritional status/risk score were not determined otherwise. Analysis of the qualitative data from the eight interviews extracted five templates with questions relevant to an increased risk of insufficient nutritional care: 1\. Divergent attitudes towards intervention. 2\. Lack of flexibility during meals. 3\. Lack of knowledge about nutritional care for patients. 4\. Nutrition -- a subordinate part of the care. 5\. Lack of recognition of responsibility for nutritional care. Divergent attitudes towards intervention ---------------------------------------- Analysis showed that the staff in the intervention sections had not been using the nutritional records systematically. Several nurses thought that the records were too comprehensive and overwhelming. Many mentioned that they had not had the time to learn how to use the records and they were clearly perceived as an extra workload. The nurses in charge mentioned that it was not unproblematic to burden staff with material they did not have the time or resources to read. However, a few staff members, among them two nurse students from bed section IIB, had learned how to use the records. They found that they were utilizable and easy to use. The two nurses in charge had divergent views on the usability of the intervention study. The charge nurse in bed section IIB thought that the intervention had improved their work with the patients\' nutrition. The staff had previously accepted that patients would lie without eating for seven to ten days. Intervention caused the staff to use a feeding tube on threatened patients earlier than before the intervention. However, the charge nurse from bed section IB declared that the staff in her section had not shown much commitment to the intervention. The staff had not taken \'ownership\' of the intervention study because the decision to participate in the project had not been a staff decision but one taken by the central management. She emphasized that the staff\'s attitude was rooted in the fact they had to take in new ideas and instructions all the time. Several care providers in bed section IB thought that it was a sizeable extra workload to use the records for recording patients\' nutritional statuses and that this had constituted a barrier to their active participation in the process. Other nurses in bed section IB declared that they did not think that it was necessary to continuously register a patient\'s nutritional status. It sufficed for some nurses to use their \'clinical judgement\' and on this basis monitor the patient\'s weight status. These nurses were not interested in any new initiatives and in tools for nutritional care. The records were not -- and are still not -- an integral part of the nutritional care in the intervention sections. This impacted on care continuity. The few staff members who had actually been using the records and had been able to identify patients at risk of insufficient nutrition reported that their observations had not been translated into action. Although the food records were only used to a minor extent, the staff generally agreed to the relevance of focusing on the patients\' nutrition. Several nurses had not previously paid much attention to the patient\'s nutrition, but the intervention had made them more conscious of this issue: \"I must say that after we have begun to pay attention to the diet, it has become clear to me how important it is. You have always known that it was important, but you do not really expect the patients to be undernourished when they are hospitalised\" Nurse After the intervention the nurses were more conscious of their choice of food rich in energy than \"before where they did not pay much attention to the fact that febrile patients constituted a special group at risk of falling into nutritional deficit\".The general belief that \'fat is bad\' for patients was widespread before the intervention. This belief springs from general dietary recommendations for healthy people. However, the intervention raised consciousness of the fact that public dietary recommendations may be suitable for healthy, but not for ill people. Lack of flexibility during meals -------------------------------- The focus group interviews overall showed that the concept of \'individual nutrition\' was not easily introduced in the nutritional practice at the two bed section. The staff was able to offer food five times during 12 hours during a 24-hour period. The duration of the meals was dictated by tight time schedules for maids and the hospital orderly. Lunch and dinner were often served under time pressure. Between the fixed meals, the care providers often lacked the time to offer patients various kinds of between-meals in the form of frozen, heated food. Assuming that nutritional care rests on the efforts of a committed staff, it may be claimed that the very organisation of the food service was counterproductive to individual nutritional care because the staff did not have real opportunities to offer the patients any food outside the fixed meal times. Individual nutritional care was also hampered by the fact that the kitchen ran a 24-hour nutrition schedule. This makes it difficult for the patients themselves to decide which meals they want to eat and hence to involve them in their own nutritional care. This was especially a problem for elderly nibblers. Lack of knowledge about nutritional care for the patients --------------------------------------------------------- The clinical dieticians disseminated knowledge about nutrition to the staff in the bed sections, e.g. knowledge about a change in diet from \'unrestricted diet\' to \'normal diet and \'hospital diet\'. However, such knowledge dissemination was obstructed in several ways. A large staff turnover in some sections meant that such knowledge did not stay in the sections. The exchange of permanent nursing staff during a seven month period including the intervention was 9% and 46% in intervention section IB and IIB respectively (Information from the administration, Aarhus University Hospital). It was difficult for the clinical dietician to get through to the entire staff, as those who were willing to listen were those who took interest in the patients\' diet: \"But those we do see are those among the staff who take active interest. It\'s the old guard turning up\" Clinical dietician Some care providers found that it was time-consuming to acquire knowledge about nutrition. Thus a member of the health care support staff mentioned that \"just learning what a \'normal diet\' and a \'hospital diet\' is takes so much time\". Along this line, several staff members, especially nursing assistants and health care support staff mentioned that it would be very useful if they had a resource agent they could ask about nutritional issues. It was possible for the care providers to refer the patients to the clinical dieticians. However, they felt that the dieticians were often so busy giving advice to referred patients that they could hardly assume a role in the daily nutritional care. The clinical dieticians, on their side, indicated that they would like the care providers to involve them more so that they could also give advice to patients who had not been referred. However, it was difficult for the clinical dieticians to be allowed to contribute: \"The nurses think that they can manage the patients\' nutritional situation. I think that is what they believe today. But if we were there when a question was raised, then they would use us. That\'s what I think\" Clinical dietician Much would be gained, according to the clinical dieticians, if the staff knew that the patients\' loss of weight during hospitalisation should be avoided and if such knowledge was used in the nutritional care. One nurse put forward the view that recommendations for healthy people also applied to patients. But during the intervention she expressed that she had changed her perception, but she found it difficult to manage nutritional requirements of ill patients and at the same time relate to dietary advice to healthy people, which were also used in the nutritional care of hospitalised patients. The clinical dieticians had also noted that many elderly patients were served the \'normal diet\' even if they needed \'hospital diet\'. These observations could signal the existence of a gap between the knowledge the nurses had and the knowledge actually needed to asses, among others, which diet suited particular patients best. The introduction of two diets caused some confusion and uncertainty among the occupational groups involved. Nutrition -- a subordinate part of the care ------------------------------------------- The care providers expressed an interest in the patients\' nutrition, but also mentioned that they often had to ignore this aspect of care because of their tight work schedule. Some days when they had the time and the resources, they would pay more attention to the patients\' nutrition, overseeing for example how much the patients were eating. But on busy days the care providers had to abstain -- e.g. from offering the patients an extra portion \"because it\'s nutrition and similar things which we must choose not to include when we are busy\". Time was a limiting factor in nutritional care. The overall message was that the staff found it difficult to find time for determining the patient\'s height, calculate BMI, talk with the patient about losing weight, the patient\'s wishes for diet and his/her possible problems with eating and drinking. Time was hence both a real and an imagined barrier to recording the patients\' nutritional status and to including the patients in their own nutritional care. In relation to that several care providers in section IB said that the nutritional care was a secondary priority. Hence, it was not perceived as a part of the care and treatment itself, but rather as a service \"along with laundering and ironing\", as mentioned by the charge nurse in section IB. Other care providers in section IB also said that serving food and beverages for the patients was not part of their job: \"You feel you are in the catering business in some way when you have to wait on the patients\" Nursing assistant This would seem to suggest that for some care providers, nutritional care and the tasks such care demanded was not perceived as a natural part of their care activities. If a part of the staff defined nutritional care and the task of making sure that the patients got enough to eat as a job function outside the normal realm of their occupation this evidently constituted a barrier to an improvement in the patients\' nutritional care. The patients\' nutrition was hence not a priority area within the overall care work performed by the nurses and it was not an active part of the treatment. Inversely, the nurse in charge in section IIB thought that nutrition should be a first line priority to ensure that the work performed by the other occupational groups could have optimal effect. She pointed, among others, to the restoration of physical strength among stroke patients. She was aware that the work routines and the barriers to knowledge dissemination to other occupational groups was a factor limiting the speed with which changes could be implemented. The nurses in charge\' attitudes to intervention and nutritional care were reflected in the attitudes of the rest of the staff. The data paradoxically showed that although nutritional care falls within the nurses\' competence area, they only engaged in such care when they had the time to do so. When the nurses were busy, which they often were according to themselves, they gave lower priority to nutritional care. However, continuity of nutritional care was particularly important in nibblers according to the clinical dietician: \"It is not always big science or intricate calculations; it\'s almost just a simple matter of remembering to serve the food to the patient\" Clinical dietician Lack of recognition of responsibility for nutritional care ---------------------------------------------------------- The results suggest that the occupational groups involved in the food service had different guidelines. The assistant catering officer from the kitchen declared that she adopted a 24-hour approach to the planning of menus and distribution of energy percentages. The care providers prioritised that patients ate the meals they chose from the menus. However, this target was compromised by constraints of time and choice. The maids stuck meticulously to the diet previously decided by the nurses for each individual patient, while the care providers did not. Moreover, contrary to the clinical dieticians, some care providers thought that overweight patients should lose weight during their admission. The dietary change introduced to make some of the patients lose weight was, however, criticized by some of care providers in section IIB. They found that the change to \'normal diet\' was not clear to the patients and was an expression of abuse of power, because the patients did not have any choice. The data gave no indications that the involved occupational groups shared a common goal as far as nutritional care was concerned. Inversely, the different groups had different priorities and showed neither insight nor any understanding of the professional competences of the other groups. The clinical dieticians also mentioned that a relatively high staff turnover at the bed sections ran counter to continuity of nutritional care and made it difficult to maintain a high, constant level of nutritional knowledge at the bed section derived through instruction and teaching undertaken by the clinical dieticians. Furthermore, the responsibility for the practical aspects of nutritional care could not be precisely located because many different staff groups were involved. This invariably increased the risk that responsibility was diluted, viz. that the individual care provider loses his/her sense of responsibility and overview of the situation. The staff in the bed sections did not see precise definition of responsibility as a central issue as opposed to staff outside the bed sections who would like to see a clear formal distribution of responsibility for nutritional care with a view to improving communication and procedures. \"Nutritional care, distribution and orders should be given priority from above. It should not be the maids who should work for this. They are often having all the problems because the care providers have other duties they must see to; so the maids are sometimes doing as best they can; and what else can they do? But it should be a priority coming from the very top\" Assistant catering officer The data suggested that dilution of responsibility was accompanied by an element of responsibility evasion. The care providers are, theoretically, responsible for the patients\' nutritional care, but the maids assumed the lion\'s share of this responsibility in practice. The maids were employed in the maintenance section and therefore had no occupational responsibility for the patients\' nutrition. However, the maids were very committed and felt responsible for the patients\' nutrition. They found it difficult to accept that they had no guarantees that other staff members would take responsibility for the patients\' diet when they were not at work. When the maids were having their weekends, holidays etc., the substitutes would often take over their function. The maids declared that they would be happy to take a more active role in the patients\' nutritional care. Through their teaching at the bed sections, the clinical dieticians had learned that the maids in general were showing much commitment, attention and responsibility towards the issue of the patients\' nutritional status. Inversely, several care providers found it difficult so see themselves take a more active role vis-à-vis nutritional care. As a measure intended to counteract the dilution and evasion of responsibility, the assistant catering officer suggested that central hospital management should issue a clear statement that the patients\' nutritional status was a high priority area that deserved serious attention from all occupational groups. Such a message could also give impetus to a process of clarifying responsibilities and tasks related to nutritional care in all bed sections. Discussion ========== Prior to intervention food ingested during hospitalisation on average met 72% of the patient\' protein (table [2](#T2){ref-type="table"}) and 85% of their energy requirement (table [3](#T3){ref-type="table"}), and there was no significant difference between the four bed sections. But the intervention targeting the nutritional care had a significantly better effect in bed section IB than in intervention section IIB measured as the extent to which the protein and energy requirements were met. But the quantitative results revealed that the forms designed for assessing the patients nutritional status had been used only to a limited extent. This result was reflected in the results showing that the staff on admission only weighed half of the patients. The outcome of the intervention was probably influenced by the reluctance among the staff in bed section IB to implement the new guidelines, and by the large staff turnover in bed section IIB. Interestingly, the patients\' intake of protein- and energy increased significant in bed section IB during the intervention. It cannot be excluded that the focus on the nutritional care coming from an investigator outside of the organization, had led to this paradox that, despite the reluctance identified among the staff, nutritional care was optimised. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Dietary intake of energy (kJ), estimated requirement for dietary energy (kJ) and degree to which requirement for energy per 24 hours was met before and after intervention. Values are group averages (standard deviation (SD)) unless otherwise stated. ::: Medical ward I Medical ward II ------------------------------------------------------------------------- ---------------- ----------------- ------------- ------------- Dietary intake of energy, kJ/24 hours (SD) 5359 (1993) 7267 (2317) 5811 (1851) 5923 (2096) Estimated mean of need for dietary energy, kJ/ 24 hours (SD) 7396 (1687) 7119 (1619) 7761 (1409) 7810 (1491) Estimated need for dietary energy per 24 hours covered in per cent (SD) 74(30) 103(24) 76(22) 76(23) 95 % confidence interval 62--87 89--116 62--89 64--89 ::: Patients who were severely mentally or physically impaired were not included in the study of ethical reasons, although they as \'nibblers\' did not receive a different form of nutritional care. So the sample is not representative for all the medical patients. If this group of patients had been included the quantitative results probably would have been lower, as described in an other Danish study \[[@B28]\]. The average length of the hospital stay for the patients participating in this study was 23 days. The average length for medical patients in the Aarhus County was six days \[[@B16]\]. This significant difference may be ascribed to the fact that patients hospitalised for less than five days were excluded in this study. On the other hand, mentally or physically impaired patients were not included. The official statistics on the length of hospital stays include a large group of patients who are long-term hospitalised. In this study 27% of the patients were hospitalised for more than four weeks. Long-term hospitalisation demands that particular attention be paid to the problem of weight loss. A 24-hour weight loss reaching 154 gram was found before the intervention, which may, indeed, be regarded as a problem during hospitalisation. The introduction of new diets made a difference both to the patients and the staff at the bed sections. Thus the \'normal diet\' had a lower fat energy percentage than the other diets. This meant that patients had to consume a very sizeable diet in order to cover their energy requirement, which was rarely manageable for patients with reduced appetite. The clinical dietician mentioned that they had most frequently met patients with a poor or reduced appetite and a simultaneous need for a diet with a high nutrient density. This observation is corroborated by observations made by other Danish clinical dieticians \[[@B29]\]. The results of this study indicate that it is hardly appropriate to base nutritional care on recommendations intended for healthy individuals if the staff\'s nutritional knowledge matches that seen in the present study. The consequences seem to be a deterioration of the nutritional status in an even larger fraction of patients. The common understanding and recognition of the integration of nutritional care as part of the overall care among all occupational groups is a key prerequisite in an effort to see nutritional care as part of the care for the individual patient \[[@B23],[@B30],[@B31]\]. Another key prerequisite is that responsibility for such care is vested in real professional competence that lies with a single staff group, i.e. that it is backed by knowledge \[[@B32]\]. On this basis it might be possible to establish cooperation and launch a fruitful dialogue. Nutritional care fell within the competence of the nurses who were therefore able largely to determine to which extent other occupational groups were allowed to contribute with knowledge about nutrition. The clinical dieticians mentioned that they would like a more extensive dialogue with the other care providers about the patients\' nutritional status, but these groups did not welcome such cooperation. One of the nurses in charge found that the intervention had made the nurses pay more attention to nutritional issues including, in particular, patients at increased risk of becoming undernourished. However, it was difficult to translate increased attention into specific nutritional care actions such as recording the patients\' nutritional status upon admission by using the special food records. In a study of the relationship between nurses\' competences and their knowledge about nutrition and diet in a hospital in the South of England, Lin Perry showed that there was no clear association between the nurses\' attitudes, knowledge and actions, as neither knowledge nor attitudes were translated into action \[[@B32]\]. The study also demonstrated discrepancies between what the nurses said they were doing in relation to the patients\' nutrition and what was actually documented in the patient records. Perry concluded that nursing care was frustrated by absence or inadequate knowledge among nurses about nutrition or by the failure to communicate such knowledge and a lack of common standards in general. However, the fact that the staff entertained views on the importance of the food and the food service did not imply that all groups were committed to seeing nutritional care as an element of the overall care effort. And the intervention study was an external project that was not anchored in the bed section\'s own staff. So that may explain the moderate reluctance to take active part in the study shown by some of the occupational groups \[[@B33]\]. The effect was that the nutritional care was not optimal. Some nurses gave as a reason for this situation that nutritional care was not part of the nursing care and that the time pressures induced by other tasks forced them to give lower priority to nutritional care. In the recommendations of The International Council of Nurses (ICN) the patients\' nutritional status is placed second after the first dimension \'ability to breathe\' \[[@B34]\]. This implies that a patient\'s nutritional status is considered an important issue in nursing theory, which is the basis of patient care. This is interesting in the light of the results of the focus group interviews presented here, because it appears that there is no agreement between the guidelines issued by the ICN and the Danish Nurses Organisation as far as the importance of nutrition and the nutritional care the patients receives during hospitalisation is concerned. Several papers in Danish and international nursing journals hence advocate that nurses assume a central role in countering patient under nourishment -- a role rarely entertained by nurses today \[[@B32],[@B35]-[@B37]\]. Yet, the clinical dieticians and the maids found that the nurses took into account neither their knowledge about patient nutrition in general, nor their knowledge of the individual patient\'s situation. Paradoxically, however, the nurses still claimed that nutritional care fell mainly within their competence. The clinical dieticians, the maids and the assistant catering officer reported poor communication between patients, nursing staff and kitchen. But the care staff did not share this view. The individual bed sections apparently did not have a clear distribution of responsibilities embracing all aspects of nutritional care. On the contrary, the data suggested that dilution of responsibility was accompanied by an element of responsibility evasion. The degree to which patients\' energy and protein requirements are covered undoubtedly varies from hospital to hospital depending on the menus served and the commitment to the nutritional care shown by the care staff and management. However, any food service involves a long chain of tasks and work processes reaching from the kitchen to the patient, and the food service is essentially organised in the same way and priorities are generally the same in all Danish hospitals. It is therefore likely that the problems associated with insufficient nutritional care are of a similar nature in Danish hospitals and some European hospitals \[[@B38]\]. The perspective for further investigation could be a Health Technology Assessment (HTA) to evaluate the aspect of the patients, the organisation and the economy of the nutritional care of medical inpatients. Conclusion ========== The average intake of energy and protein among hospitalised medical patients did not cover their requirements. Prior to intervention, food ingested during hospitalisation on average met 72% of the patients\' protein and 85% of their energy requirement. After changing the diets from \'unrestricted diet\' to \'normal diet\' and \'hospital diet\', the diet on average met 61% of the control patients\' protein and 75% of their energy requirements. Intervention allowed a significantly better satisfaction of the patients\' protein and energy requirements at one of the intervention sections using standard hospital food. However, the implementation of procedures focusing on nutritional care appeared to be difficult, especially at bed sections with a large staff turnover. Consequently, the results of the study call attention to the existence of barriers to efforts aimed at improving the nutritional care of patients. Introduction of nutritional care as part of the overall care met with barriers among the care providers. Focus group interviews identified these barriers as lack of time, lack of knowledge, lack of contact with resource agents concerning nutrition, lack of commitment, resistance towards a additional perceived workload and resistance towards providing service to the patients. Care providers who wished to provide individual nutritional care saw the very organisation of the food service as an obstacle to their freedom of action and flexibility. The effect of this was that it was difficult to accommodate individual patients\' requirements. Occupational groups involved in nutritional care worked on the basis of different perceptions, had no shared target and no clear division of responsibility. Improvement of the nutritional care requires that focus be directed towards the final link in the food service chain. This study showed that nutritional care was a subordinate rather than a coordinate element in the overall care effort. The failure of coordination hinged on dimensions of organisation, knowledge and resource utilisation and it significantly affected the degree to which patients\' nutritional requirements were met. An increase in the priority given to nutritional care by central hospital management and a concomitant general change in attitude towards nutritional care is needed and is probably a precondition for achieving a level of sufficient nutrition among hospitalised patients. Authors\' contributions ======================= Karin O. Lassen carried out the research design, the fundraising, coordination of organisational communication, food record planning and implementation, performed the data analysis, drafted the manuscript and is the guarantor of the manuscript. Filip Kruse participated in the designing the focus group interviews, in the data analysis, and as author of the manuscript. Merete Bjerrum participated in the data analysis and as author of the manuscript. Lillian Jensen participate in the planning of the food records, performed the calculation of protein and energy intake and participate in the discussion of the manuscript. Kjeld Hermansen contribute to organisational support and discussion of research design and manuscript. All authors read and approved the final manuscript. Competing interests =================== None declared. Acknowledgements ================ Thanks are due to patients and staff at the kitchen and the participating wards, Aarhus University Hospital, Denmark, to the statistical support from the Western Danish Research Forum for Health Sciences, and to The Ministry of Health, The Health Insurance Foundation and The Foundation of Aase and Ejnar Danielsen, Denmark for sponsorship.	PubMed Central	2024-06-05T03:55:48.361873	2004-9-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521697/", "journal": "Nutr J. 2004 Sep 13; 3:12", "authors": [ { "first": "Karin O", "last": "Lassen" }, { "first": "Filip", "last": "Kruse" }, { "first": "Merete", "last": "Bjerrum" }, { "first": "Lillian", "last": "Jensen" }, { "first": "Kjeld", "last": "Hermansen" } ] }
PMC521727	If you ask the average ten year old in America what a scientist looks like, they almost always describe an older man with crazy white hair and a lab coat. If you ask a group of adolescents how many have looked through a microscope, few raise their hands. If you discuss the implications of genetic research with a group of high school students, they\'re likely to cut your next class. The reason why these students have such profound stereotypes of scientists and are less than enthusiastic about science\'s impact on society is simple---the lack of exposure they receive during their pre-college education. According to a preliminary study conducted at Leicester University in England, students are often repeatedly confronted with stereotypes of science and scientists via television, cartoon, and comic book characters as well as uninformed adults or peers ([@pbio-0020279-McDuffie1]). A university set in a major city has the resources to change the mindset of urban students and engage them in the exciting field of science. At Thomas Jefferson University (TJU), located in Philadelphia, Pennsylvania, a team of scientists and educators has developed a program that breaks down the stereotypes of the science field and allows students to engage in real, live experiments at their own schools. Why Are Such Programs Necessary? {#s2} ================================ Around the world, educators face difficult choices in focusing educational goals with limited resources. In the United States, the No Child Left Behind Act, which guides school funding policy, currently places an emphasis on literacy and math in schools, with the result that best programs and practices in education are increasingly directed toward these two areas. Unfortunately, science education has become a lesser priority. Teachers are not given adequate resources to allow students to "get their hands dirty" during science lessons. If we want a society that is interested and knowledgeable about the need for scientific research, the basic principles of the life sciences need to be integrated early in the pre-college curriculum ([@pbio-0020279-Sylwester1]). We developed the Thomas Jefferson University Science Outreach Program (TJUSOP) to address this key issue and provide inquiry-based educational strategies through collaborative efforts between the university\'s faculty and partnering school districts. How Are We Making a Difference? {#s3} =============================== Using the TJU facilities and laboratories, this innovative program integrates life science into the education of students between the ages of eight and eighteen from Pennsylvania, New Jersey, and Delaware. The mission of the TJUSOP is to foster an enthusiasm for science education, promote interest in future participation in biology-related fields, and allow all students the opportunity to learn life science through a hands-on, student-centered approach to instruction. The program is a supplement to the established curriculum, developed to support the content knowledge that is taught at each grade level. Teachers are invited to attend a professional development workshop held at the beginning of the school year where they receive training and resources for the units. Then, TJUSOP educators assist the teachers and students in their own classroom in running a weeklong experiment. This allows a large amount of group work to be completed simultaneously, even when teachers are faced with time constraints and large class sizes. This program is at no cost to the districts participating and is funded through the Jefferson Medical College and the Kimmel Cancer Center, as well as through the generosity of several local and national groups including Glaxo Smith Kline, the Christopher Ludwick Foundation, the Joan and Joseph Fernandez Family Foundation, the Brook J. Lenfest Foundation, the Foerderer Foundation, Drinker Biddle and Reath, and the Pennsylvania Department of Agriculture. Since its inception in August of 2002, this program has reached over 2,000 students and 75 teachers through our one-week zebrafish classroom experiments, our hands-on zebrafish and Drosophila facility tours in conjunction with Dr. James Jaynes (a Drosophila scientist at the Kimmel Cancer Center), and our High School Mentorship program held each summer. One of the main goals of TJUSOP is to reach students from ethnic and economic groups that are underrepresented in the scientific community. We have successfully partnered with the School District of Philadelphia, where 84.9% of the students come from ethnic backgrounds other than Northern European and 80% of the students are eligible for free or reduced-cost meals. This district, among many others, receives pre- and post-instruction for all teachers and students at no cost to the district, allowing it to improve the quality of its science education. Although we target these school districts, it is important to note that the US is facing a problematic decrease in the number of Americans, of any background, entering the science and engineering workforce. According to a National Science Foundation report, "If action is not taken now to change these trends, we could reach 2020 and find that the ability of U.S. research and education institutions to regenerate has been damaged and that their preeminence has been lost to other areas of the world" ([@pbio-0020279-NSBC1]). In regard to this unsettling discovery, TJUSOP welcomes all districts to participate and hopes to secure funding to double the number of students reached per year. Anyone Can Be a Scientist {#s4} ========================= Our pedagogical approach to experiments allows students and teachers to become scientists, following the scientific process from beginning to end. Our live, one-week classroom experiments for the fourth, seventh, and tenth grades use zebrafish, a popular model organism for genetic research. A curriculum sample is as follows: in the seventh-grade unit, the students mate albino (recessive trait) male and a wild-type (dominant trait) female zebrafish in order to observe what the offspring will look like. Students form hypotheses, such as that the young offspring will look like the mother and the older offspring will be striped. Throughout the week, students observe and record embryos developing a head, tail, and notochord and pigment development. By the end of the experiment, a live heartbeat can be seen as well as the individual blood cells flowing throughout the larvae using a stereomicroscope TJUSOP provides ([Figure 1](#pbio-0020279-g001){ref-type="fig"}). ::: {#pbio-0020279-g001 .fig} Figure 1 ::: {.caption} ###### A Philadelphia Student Uses the Zeiss Stereomicroscope during the Weeklong Experiment ::: ![](pbio.0020279.g001) ::: Grade-specific scientific journals are given to the students. The journals contain an introduction to TJUSOP and the experiment, background information about zebrafish in research, scientific vocabulary words used throughout the unit, and a word search activity. Students are given the title of "Junior Scientists" in grades 4 and 7 and "Student Scientists" in grade 10 and are asked to record the research question, a hypothesis, daily observations, and the conclusion of the experiment ([Figure 2](#pbio-0020279-g002){ref-type="fig"}). ::: {#pbio-0020279-g002 .fig} Figure 2 ::: {.caption} ###### An Example of a Seventh-Grade Journal Entry ::: ![](pbio.0020279.g002) ::: How to Break the Stereotypes of What Science Is {#s5} =============================================== TJUSOP allows student participants to use scientific tools, talk with real scientists, and gain scientific knowledge so they can become informed members of their communities. Upon asking a fourth-grade student why she thought it was important to learn about science using zebrafish and the microscope, the student wrote, "I think it is important because we can find facts about oursefs \[sic\]." This sounds like a good start. For more information about the program, or if you would like to get involved in the initiative, please contact Jamie Schaefer, at jamie. E-mail: <Schaefer@mail.jci.tju.edu> or visit <http://www.kimmelcancercenter.org/scienceoutreachprogram>. Jamie Schaefer and Steven Farber are at the Kimmel Cancer Center, Thomas Jefferson University, Philadelphia, Pennsylvania, United States of America. TJU : Thomas Jefferson University TJU-SOP : Thomas Jefferson University Science Outreach Program	PubMed Central	2024-06-05T03:55:48.371588	2004-10-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521727/", "journal": "PLoS Biol. 2004 Oct 12; 2(10):e279", "authors": [ { "first": "Jamie", "last": "Schaefer" }, { "first": "Steven A", "last": "Farber" } ] }
PMC521728	Diatoms, unicellular algae with ornate silica shells, have fascinated amateur and professional biologists ever since the invention of the microscope. But these days, diatoms and their exquisite shells are also attracting the attention of nanotechnologists who hope that diatoms will teach them how to make minute structures currently beyond the capabilities of materials scientists. And now these nanotechnologists, together with ecologists interested in the global carbon cycle---in which diatoms play a central role---have a genomic blueprint to help them in their studies: the annotated genome sequence of Thalassiosira pseudonana (<http://genome.jgi-psf.org/diatom/>). What Are Diatoms? {#s2} ================= Diatoms, microalgae that are found in all aquatic and moist environments, first appeared more than 180 million years ago. Since then, diatom diversity has literally exploded; no one is sure how many living species there are---probably about 100,000---or why there are so many different types. Plant molecular biologist Chris Bowler (Ecole Normale Supérieure, Paris, France and Stazione Zoologica, Napoli, Italy) explains that molecular phylogeny and morphological studies suggest that diatoms originated 'probably as the result of a eukaryote being invaded or engulfed by a photosynthetic eukaryote, most probably a red alga'. The basic structure of all diatoms is similar: a single cell, often with a large vacuole, contained within a silica shell or frustule made of two overlapping halves or valves joined by girdle bands, which are also made of silica. The girdle bands form the rims of the two valves and allow unidirectional growth of the diatom during vegetative division. 'The shell is rather like a Camembert cheese box or a petri dish', explains marine ecologist Christian Hamm (Alfred Wegener Institute for Polar and Marine Research, Bremerhaven, Germany). There are only two main types of diatom: centric diatoms, which often have a circular symmetry, and pennate diatoms, which are usually bilaterally symmetrical. Nevertheless, diatom shells come in a dazzling array of forms and sizes ([Figure 1](#pbio-0020306-g001){ref-type="fig"}; [Box 1](#box1){ref-type="boxed-text"}). 'They can be circular, oval, stick-shaped, you name it, and range from several micrometres large to about a millimetre', says ecologist Mary Ann Tiffany (San Diego State University, California, United States), who is using scanning electron microscopy to examine diatom valve formation as part of her graduate studies. 'When a diatom divides, each daughter cell makes a new half shell', explains Tiffany. The first stage of construction is the generation and deposition of silica nanospheres; the more ornate structures are built up from there. Both the finished shells, with their precise and reproducible nanometre-scale features, and the intermediate structures that lead up to the finished product, could be of interest to nanotechnologists, suggests Tiffany. ::: {#pbio-0020306-g001 .fig} Figure 1 ::: {.caption} ###### Scanning Electron Micrographs of Diatoms \(A) Biddulphia reticulata. The whole shell or frustule of a centric diatom showing valves and girdle bands (size bar = 10 micrometres). \(B) Diploneis sp. This picture shows two whole pennate diatom frustules in which raphes or slits, valves, and girdle bands can be seen (size bar = 10 micrometres). \(C) Eupodiscus radiatus. View of a single valve of a centric diatom (size bar = 20 micrometres) \(D) Melosira varians. The frustule of a centric diatom, showing both valves and some girdle bands (size bar = 10 micrometres). (Images courtesy of Mary Ann Tiffany, San Diego State University.) ::: ![](pbio.0020306.g001) ::: Turning to Nature for Engineering Solutions {#s3} =========================================== Richard Gordon, Professor of Radiology at the University of Manitoba in Winnipeg, Canada, somewhat accidentally laid the foundations of 'diatom nanotechnology' in 1988 when he was invited to give a lecture at an engineering conference. 'I\'m not an engineer', explains Gordon, 'but I knew engineers were interested in what was then called microfabrication so I told them about diatoms because they are so good at making small things'. Gordon, a keen diatom hobbyist, explained to his audience how diatoms could make a three-dimensional micro- or nanoscale structure for them without them lifting a finger. By contrast, says Gordon, 'nanotechnology techniques then and now are tedious, involving painstakingly building three-dimensional structures up layer by layer'. Such tedious techniques are currently used in the semiconductor industry. At present, explains Michael Sussman, Director of the Biotechnology Center at the University of Wisconsin-Madison (Madison, Wisconsin, United States), 'features are etched onto circuit boards using light. However, the wavelength of light limits the smallest size that can be achieved, and for the next generation of faster computers, engineers need to get denser features onto computer chips than is possible with light etching'. Diatoms, says Sussman, 'are natural-born lithographers in the nanometre range. If we could work out how diatoms lay down micro lines of silica, then we may be able to simulate it'. The proteins that diatoms use to direct silica deposition could be very useful to the semiconductor industry, says Sussman. There are other ways in which diatoms could help us clumsy humans build nanoscale 'widgets'. Molecular biologist Mark Hildebrand (Scripps Institution of Oceanography, San Diego, California, United States) is a member of a collaborative project trying to develop genetically engineered micro/nanodevices (also called GEMs). Already, engineers are using diatoms to help them build extremely sensitive sensors based on microfluidic devices, he explains. Hildebrand is also interested in the optical properties of diatoms. 'Information processing technology is moving from electronically to optically based hardware, which allows more information to be carried and stored. Optical systems need materials with regularly repeating structures with features below the micrometre size range. These are very difficult to make by standard manufacturing techniques, but diatoms make structures like this all the time'. It might also be possible to use diatom shells as delivery vehicles for drugs, suggests chemical engineer Tony Rogers, an assistant professor at Michigan Technological University (Houghton, Michigan, United States). 'They have a uniform nanoscale pore structure and are chemically inert and biocompatible'. Rogers envisages loading diatoms with a drug that would then leach out into the blood stream at a rate dependent on the diatom species used. By incorporating ferromagnetic particles within the diatom structure, it might be possible to use a magnet to guide the drug to the right organ, he suggests. Diatom structures are not just of interest to people interested in tiny objects. As Hamm comments, 'in diatoms, Nature has solved many of the problems that engineers want to solve. For example, diatoms are particularly good at making lightweight but strong structures. Because it is possible to scale static structures like shells, diatoms can teach us how to make lightweight constructions for the aerospace and car industry'. Some of the potential applications of diatoms can be investigated right now, using naturally occurring diatoms. In addition, subtle but important changes can be induced in diatoms by varying the amount of silica in their environment or changing the water flow. Gordon also envisages a device he calls a compustat, which would be used to select diatoms for a specific purpose. Diatoms taken from the sea, for example, would be individually examined using a computer-controlled microscope. 'We would tell the computer what characteristics we were looking for, and it would go through the culture, zapping those diatoms furthest from the ideal with a laser beam. The culture would then be allowed to grow up again and the process repeated until we got the sort of diatoms we wanted', says Gordon. Gordon has not built a compustat yet---it may not work, he says, because we don\'t know how far we can push diatoms by forced evolution. And even if the compustat does work, to make the most of the nanotechnological potential of diatoms, we need to know exactly how diatoms make their shells. At present, all we know is that silicon transporters and a group of long-chain, polyamine-containing proteins called silaffins, which act as nucleation points for silica deposition, are involved. This is where the diatom sequencing project at the United States Department of the Environment\'s Joint Genome Initiative (JGI) at Walnut Creek, California, comes in. The First Diatom Is Sequenced {#s4} ============================= Daniel Rokhsar, Department Head for Computational Genomics at JGI, explains why his institute undertook the sequencing and computer annotation of the genome of T. pseudonana, a marine centric diatom 'We believe that knowing this genome will help us to figure out how to mimic the processes that diatoms use to construct their very precise structures, and that we can then learn how to create similarly precise structures ourselves'. Also, he adds, diatoms are extremely important on an ecological level. Oceanographer Ginger Armbrust (University of Washington, Seattle, Washington, United States), Principal Investigator on the sequencing project for T. pseudonana, explains further. Diatoms are responsible for between 25% and 40% of all the primary productivity of the oceans, she says. 'They also keep the biological pump going. By fixing carbon dioxide and then sinking, diatoms draw carbon dioxide out of the atmosphere and take it into the deeper waters of the ocean, where it is retained for longer than it would be if the diatoms stayed near the surface'. T. pseudonana, she continues, was chosen as the first diatom to sequence in part because it has a small genome, but mainly because it represents a cosmopolitan genus of diatoms and its physiology has been well studied. Once the primary sequence of the genome had been determined, molecular biologists, oceanographers, and ecologists from around the world gathered at JGI for a 'genome jamboree'. 'The first of these was in October 2002, a massive brainstorming session at which we all dug around in the genome for our favourite genes and tried to get a feel for what was there', explains Bowler. 'It was really refreshing to get the insights of oceanographers and ecologists into what this genome was telling us'. Among other things, Armbrust and her collaborators are interested in finding out what the T. pseudonana genome can tell them about the difference between photosynthesis on land and in the sea. They also want to investigate how these organisms adapt to their environment. 'Now that we have the genome', says Armbrust, 'we can investigate how gene expression varies at different places in the water column, for example. This will be the first time a eukaryotic genome has been interpreted in this ecological sort of way'. What About Silicon Metabolism and the Nanotechnology Dream? {#s5} =========================================================== 'One of the striking things about the T. pseudonana genome is that we can figure out quite a bit from it about how this diatom deals with organic materials, but it is hard to figure out what it is doing with silicon', admits Rokhsar. 'The only way we can really figure out what a gene is doing is by comparing it with known genes in other organisms, but because diatoms are so unique in their use of silicon, we don\'t have that option. We literally just have the parts list'. To get a hook on which of the 10,000 or so T. pseudonana genes is important in silicon metabolism, Sussman is using microarrays to investigate how silicon concentrations affect gene expression patterns in the diatom. 'There may be a few hundred genes whose expression changes in response to silicon stress', he predicts, 'and we can then focus on the role that these genes play in silicon metabolism'. In another approach, Hildebrand is purifying the proteins present in diatom shells. 'Once we have isolated these proteins, we can get a little bit of protein sequence, and from there go back to the genome to pull the gene out', he explains. In an ideal world, the next step would be to see what effect genetically altering the expression of the proteins identified by Sussman and Hildebrand has on the silica shell of T. pseudonana. Unfortunately, this can\'t currently be done. 'The only diatom we can genetically manipulate is Phaeodactylum tricornutum, a pennate diatom', explains Bowler. P. tricornutum, he says, is the 'lab rat' of the diatom world but is much less important ecologically than T. pseudonana ([Figure 3](#pbio-0020306-g003){ref-type="fig"}). Bowler has previously determined the size of P. tricornutum\'s genome and is now leading a JGI project that is 75% of the way through sequencing the P. tricornutum genome. 'It will be critical to have this second genome', notes Rokhsar, 'because it will highlight what is unique to this group of organisms, and provide additional help in pulling out silicon metabolism genes'. ::: {#pbio-0020306-g003 .fig} Figure 3 ::: {.caption} ###### Light Micrograph of Phaeodactylum tricornutum This pennate diatom is the 'lab rat' of diatoms, and its genome sequence is currently being determined. (Image courtesy of Alessandra de Martino and Chris Bowler, Stazione Zoologica and Ecole Normale Supérieure.) ::: ![](pbio.0020306.g003) ::: Once the details of silicon metabolism have been revealed, the stage should be set for nanotechnologists to harness diatom proteins for the manufacture of nanodevices. 'Whether we use those proteins inside the diatom or in test tubes remains to be seen, but one way or another, diatoms are harbouring a secret that engineers need to learn about', says Sussman. Hildebrand agrees, noting how 'important it is that materials scientists recognise the incredible ability of biology to make structures that could perhaps be incorporated in the design of nanotechnological widgets'. For Armbrust, it is the ecological insights are coming out of the T. pseudonana genome sequencing project---which is part of a bigger JGI program on algal genomics---that are most exciting. 'Already, multiple little insights are encouraging us to think differently about how diatoms perceive their environment and survive in it. We have also seen many things we can\'t figure out at all right now. My heart lies in the ecology of these organisms, but if we can generate information that leads to spinoffs for nanotechnology, that will be fantastic', she concludes. ::: {#pbio-0020306-g002 .fig} Figure 2 ::: {.caption} ###### Diatoms in Art (Image courtesy of David Roberts, University of Wales, Bangor, UK.) ::: ![](pbio.0020306.g002) ::: ::: {.caption} ###### Box 1. Diatoms in Art ::: Diatoms don\'t inspire only biologists and engineers---artists, too, are fascinated by their intricate structures. Exquisite line drawings produced by zoologist Ernst Haeckel influenced the Art Nouveau movement, and more recently, wood-worker Louise Hibbert and jeweller Sarah Parker-Eaton have collaborated to produce three-dimensional objects based on diatoms and other plankton. 'We both independently used marine biology as a source of inspiration for our art', explains Parker-Eaton. 'As a student, I often visited the Natural History Museum in London, where there were drawers of fascinating marine organisms that I could sketch'. In January 2002, Hibbert and Parker-Eaton were invited to the marine laboratories at the University of Wales at Bangor (United Kingdom) by oceanographer David Thomas. 'What we saw down the microscopes just blew our socks off', says Parker-Eaton. 'We could see how the plankton moved, the forms, the incredible different layers within the diatoms. What we particularly liked about diatoms was their complexity---they are totally unlike any other life form'. Parker-Eaton and Hibbert translate what they see down the microscope into objects made of wood and silver, usually small enough to hold in the hand. [Figure 2](#pbio-0020306-g002){ref-type="fig"} shows a representation of Navicula sp. The main body is sycamore, the 'blobs' are resin, and the object is coloured with inks. The piece is held together by magnets but splits in half to reveal two silver inserts at its centre. More examples of these artists\' work can be seen at <http://www.louiseandsarah.com>. Jane Bradbury is a freelance science news writer based in Cambridge, United Kingdom. E-mail: <janeb@sciscribe.u-net.com> JGI : Joint Genome Initiative	PubMed Central	2024-06-05T03:55:48.372919	2004-10-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521728/", "journal": "PLoS Biol. 2004 Oct 12; 2(10):e306", "authors": [ { "first": "Jane", "last": "Bradbury" } ] }
PMC521729	Sitting in the enveloping quietness of an anechoic chamber, or other quiet spot, you soon become aware that the ear makes its own distinctive sounds. Whistling, buzzing, hissing, perhaps a chiming chorus of many tones---such continuous sounds seem remarkably nonbiological to my perception, more in the realm of the electronic. Even more remarkable, put a sensitive microphone in the ear canal and you will usually pick up an objective counterpart of that subjective experience. Now known in auditory science as spontaneous otoacoustic emission, the sound registered by the microphone is a clear message that the cochlea uses active processes to detect the phenomenally faint sounds---measured in micropascals---that our ears routinely hear. If the ear were more sensitive, we would need to contend with the sound of air molecules raining upon our eardrums. What is that process---the mechanical or electrical scheme that Hallowell Davis in 1983 called the 'cochlear amplifier' ([@pbio-0020337-Davis1])---which energises the pea-sized hearing organ buried in the solid bone of our skull? That question has engaged my curiosity since the late 1970s, when English auditory physicist David Kemp first put a microphone to an ear and discovered the telltale sounds of the cochlea at work ([@pbio-0020337-Kemp1]). Siren-like, the sounds have drawn me into the theory and experiment of cochlear mechanics, now as part of a PhD course at the Australian National University in Canberra. I am studying the micromechanics of this process from a theoretical point of view, and investigating whether a resonance picture of some kind can be applied to the faint but mysterious sounds most cochleas emit. Kemp\'s discoveries are rightly viewed as opening a fresh path to auditory science, and to the tools and techniques for diagnosing the functional status of the cochlea. But in terms of fundamental understanding, a key paper remains that of Thomas Gold more than half a century ago ([@pbio-0020337-Gold1]). Still cited widely today, this paper deals with the basic question of how the cochlea works to analyse sound into its component frequencies. Two prominent theories---sympathetic resonance, proposed by [@pbio-0020337-Helmholtz1], and travelling waves, proposed by [@pbio-0020337-vonBekesy1]---need to be distinguished ([Figure 1](#pbio-0020337-g001){ref-type="fig"}). In a nutshell, are there tiny, independently tuned elements in the cochlea, like the discrete strings of a piano, that are set into sympathetic vibration by incoming sound (Helmholtz), or is the continuously graded sensing surface of the cochlea hydrodynamically coupled so that, like flicking a rope, motion of the eardrum and middle ear bones causes a travelling wave to sweep from one end towards the other (von Békésy)? ::: {#pbio-0020337-g001 .fig} Figure 1 ::: {.caption} ###### Two Views of Cochlear Mechanics The cochlea, shown uncoiled, is filled with liquid. In the accepted travelling wave picture (A), the partition vibrates up and down like a flicked rope, and a wave of displacement sweeps from base (high frequencies) to apex (low frequencies). Where the wave broadly peaks depends on frequency. An alternative resonance view (B) is that independent elements on the partition can vibrate side to side in sympathy with incoming sound. It remains open whether the resonant elements are set off by a travelling wave (giving a hybrid picture) or directly by sound pressure in the liquid (resonance alone). ::: ![](pbio.0020337.g001) ::: The first option, sympathetic resonance, has the advantage of allowing vanishingly small energies to build up, cycle by cycle, into an appreciable motion---like boosting a child on a swing. The second, travelling wave, has the weight of von Békésy\'s extensive experiments behind it. At the same time, one of the drawbacks of the travelling wave theory is the difficulty of accounting for the ear\'s exquisite fine tuning: trained musicians can easily detect tuning differences of less than 0.2%. Even von Békésy himself notes, on page 404 of his classic book, that 'the resonance theory of hearing is probably the most elegant of all theories of hearing'. Gold\'s work, done in collaboration with RJ Pumphrey ([@pbio-0020337-Gold1]), was the first to consider that the ear cannot act passively, as both Helmholtz and von Békésy had thought, but must be an active detector. Gold was a physicist who had done wartime work on radar, and he brought his signal-processing knowledge to bear on how the cochlea works. He knew that, to preserve signal-to-noise ratio, a signal had to be amplified before the detector, and that 'surely nature can\'t be as stupid as to go and put a nerve fibre---that is a detector---right at the front end of the sensitivity of the system'. He therefore proposed that the ear operated like a regenerative receiver, much like some radio receivers of the time that used positive feedback to amplify a signal before it was detected. Regenerative receivers were simple---one could be built with a single vacuum tube---and they provided high sensitivity and narrow bandwidth. A drawback, however, was that, if provoked, the circuit could 'take off', producing an unwanted whistle. Gold connected this with the perception of ringing in the ear (tinnitus), and daringly suggested that if a microphone were put next to the ear, a corresponding sound might be picked up. He experimented, placing a microphone in his ear after inducing temporary tinnitus with overly loud sound. The technology wasn\'t up to the job---in 1948 microphones weren\'t sensitive enough---and the experiment, sadly, failed. Gold\'s pioneering work is now acknowledged to be a harbinger of Kemp\'s discoveries. But there is one aspect of Gold\'s paper that is not so widely considered: Gold\'s experiments led him to favour a resonance theory of hearing. In fact, the abstract of his 1948 paper declares that 'previous theories of hearing are considered, and it is shown that only the resonance theory of Helmholtz... is consistent with observation'. Gold and Pumphrey did psychophysical experiments in which hearing thresholds were determined for listeners first for continuous pure tones and then for increasingly briefer stimuli of the same frequency. Gold and Pumphrey showed that their results could only be accounted for by considering the cochlea as a set of resonators, each of which responds to a narrow frequency range. In a second neat experiment, listeners had to detect differences between the sound of repetitive tone pips (series one) and those same stimuli but with the phase of every second pip inverted (series two, in which compressions replaced rarefactions and vice versa). Out-of-phase pips should counteract the action of in-phase pips and, following the child-on-swing analogy, rapidly bring swinging to a halt. Therefore, the argument goes, the two series should sound different. By increasing the silent interval between pips until the difference disappeared, the experimenters could infer how long the vibrations (or swinging) appeared to persist and could then put a measure on the quality factor (Q), or narrowness in frequency range, of the presumed underlying resonance. From the first experiment, Gold and Pumphrey derived values of Q of 32 to 300, meaning that the range of response was as little as 1/300-th of the imposed frequency---based on the picture of a broad travelling wave. The second experiment gave comparable results. However, their resonance interpretation has been dismissed because of a methodological flaw in the second experiment: the spectral signatures of the two series are not the same and provide additional cues. Nevertheless, it is not widely appreciated that the first experiment seems methodologically sound, and its results remain persuasive. I think the resonance theory deserves reconsideration. The evidence of my ears tells me that the cochlea is very highly tuned, and an active resonance theory of some sort seems to provide the most satisfying explanation. Furthermore, as well as Gold\'s neglected experiment, we now know from studies of acoustic emissions that the relative bandwidth of spontaneously emitted sound from the cochlea can be 1/1000 of the emission\'s frequency, or less. My research, guided by Professors M. V. Srinivasan and N. H. Fletcher, has centred on finding an answer to that most fundamental question: if the cochlea is resonating, what are the resonant elements? A point of inspiration for me is Gold\'s later discussion of cochlear function ([@pbio-0020337-Gold2])---some nine years after Kemp\'s discoveries had been made. Gold draws a striking analogy for the problem confronting the cochlea, whose resonant elements---whatever they are---sit immersed in fluid (the aqueous lymph that fills the organ). To make these elements resonate is difficult, says Gold, because they are damped by surrounding fluid, just like the strings of a piano submerged in water would be. He concludes that, to make 'an underwater piano' work, we would have to add sensors and actuators to every string so that once a string is sounded the damping is counteracted by positive feedback. 'If we now supplied each string with a correctly designed feedback circuit,' he surmises, 'then the underwater piano would work again.' My research is investigating what Gold\'s underwater piano strings might be. A suggestion put forward in a recent paper ([@pbio-0020337-Bell1]) is that resonance might occur in the space between the cochlea\'s geometrically arranged rows of outer hair cells. These cells are both effectors (they change length when stimulated) and sensors (their stereocilia detect minute displacements), so a positive feedback network can form that sets up resonance between one row of cells and its neighbour. The key is to transmit the feedback with the correct phase delay, and the new paper describes how this can be done using 'squirting waves' in the gap occupied by the outer hair cell stereocilia. The paper suggests that the outer hair cells create a standing wave resonance, from which energy is delivered to inner hair cells (where neural transduction takes place). In this way, the input signal is amplified before it is detected---an active system functioning just like Gold\'s regenerative receiver. With a prime candidate in place for the resonating elements, this should, I think, prompt us to re-evaluate resonance theories of hearing, which were first put forward by the ancient Greeks and which, irrepressibly, keep resurfacing. The best-known resonance theory was that formulated by Helmholtz, but at that time no satisfactory resonating elements could be identified, and it lapsed until Gold\'s attempt to revive it. There are other difficulties in reviving a resonance theory of hearing, but I think they can be overcome. If there really are resonant elements in the ear, the outstanding question would be, how are they stimulated? It is conceivable that motion of the conventional travelling wave sets them off, in which case we have an interesting hybrid of travelling wave and resonance. The other possibility, which I favour, is that outer hair cells are stimulated by the fast pressure wave that sweeps through all of the cochlear fluid at the speed of sound in water (1,500 m/s). If that is the case, and outer hair cells are primarily pressure sensors, not displacement detectors, then the ear is a fully resonant, pressure-driven system. New life, perhaps, to that old resonance idea. AB is supported by a PhD scholarship from the Australian National University. Andrew Bell is a graduate student in the laboratory of Mandyam Srinivasan at the Research School of Biological Sciences at the Australian National University, Canberra, Australia. E-mail: <andrew.bell@anu.edu.au>.	PubMed Central	2024-06-05T03:55:48.374755	2004-10-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521729/", "journal": "PLoS Biol. 2004 Oct 12; 2(10):e337", "authors": [ { "first": "Andrew", "last": "Bell" } ] }
PMC521730	A ban in the 1866s by the French Academy of Sciences on publications about the origin of human language must have been one of the strangest bans in the history of sciences. Yet it was highly effective. After the ban, scientists and interested laymen had to wait for more than a century to hold a textbook on language evolution in their hands. Language Evolution, a compilation of essays by a diverse group of respected researchers, is amongst the first books that try to tackle what is arguably one of the hardest scientific problems. The editors set themselves the ambitious target of creating an up-to-date book about this emerging field, and they have to be congratulated for their efforts. Linguists, cognitive scientists, behavioural ecologists, and theoretical biologists all offer their view on the origin of human language and, refreshingly, do not shy from pointing out the real or assumed weaknesses of the other approaches.[](#pbio-0020346-g001){ref-type="fig"} ::: {#pbio-0020346-g001 .fig} ![](pbio.0020346.g001) ::: One of the main themes of the book is the evolutionary approach and the importance of biological structures and properties that were co-opted in the development of language (pre-adaptations). In one essay, Michael Studdert-Kenedy and Louis Goldstein propose that speech, as a motor function, draws on phylogenetically ancient mammalian oral capacities for sucking, licking, swallowing, and chewing. Thus, our hominid ancestors adopted an apparatus already divided neuroanatomically into discrete components. Complementing this evidence, Marc Hauser and Tecumseh Fitch compare human speech production and perception with that of nonhuman species. They conclude that many traits that were formerly thought to have evolved specifically for speech (such as having a descended larynx or categorical perception) are also present in other species. But perhaps the most interesting idea about pre-adaptation comes from the work of neuroscientist Michael Arbib on 'mirror' neurons in monkeys. These neurons are a subset of the grasp-related premotor neurons that discharge not only, as other premotor neurons do, when the monkey executes a certain class of actions, but also when the monkey observes more or less similarly meaningful hand movements made by the experimenter (or by another monkey). The area in which these grasp-related neurons are found is analogous with the Broca\'s area in human brains, which is involved in assessing the syntax of words. This observation serves as the basis for the mirror-system hypothesis, which postulates that Broca\'s area in humans evolved from a basic mechanism not originally related to communication but rather from the mirror system for grasping in the common ancestor of monkey and human. As a result, the mirror system provides a possible 'neural link' in the evolution of human language. There is still much debate about the selection pressures that led to the evolution of language. Observing the overabundance of potential selective scenarios for why language evolved, the linguist Derek Bickerton voices his scepticism: 'The fact that these and similar explanations flourish side by side tells one immediately not enough constraints are being used to limit possible explanations.' One frequent source of confusion, he notes, is equating language with speech by not distinguishing between modality, lexicon, and structure. Hauser and Fitch share Bickerton\'s scepticism and urge scientists to rely more on the traditional comparative approach, which was always the strength of Darwinian evolutionary theory. Primatologist Robin Dunbar, who originally proposed that grooming (group bonding) could have provided the stimulus for language, dismisses two other possible scenarios---hunting and tool-making---as potential ecological contexts for the evolution of human language. Gestural origins are also dismissed in his theory, because gestural languages do not seem to develop spontaneously and also require a line-of-sight contact making them useless at night. Interestingly, Steven Pinker rules out both Dunbar\'s theory of grooming and Geoffrey Miller\'s theory of sexual selection, whereas Bickerton rules out grooming, gossip, mating contract, and Machiavellian intelligence as likely contexts for the origin of human language. Also under fire in the book is the idea that the human brain is somehow equipped at birth with a 'universal grammar' out of which all human languages later develop. Several authors try to provide alternatives to innate predispositions, such as the importance of function to categorization (Michael Tomasello) and the importance of cultural transmission to the structure of language (Simon Kirby and Morton Christiansen). Arbib explicitly questions the traditional Chomskyan theory of innate linguistic predispositions and argues that what humans have and had in the past is 'language readiness' rather than a fixed universal grammar. Neuroscientist Terrence Deacon also puts an alternative theory forward. According to Deacon, many of the language universals reflect semiotic constraints inherent in the requirements for producing symbolic reference rather than innate predispositions. Thus, neither evolved innate predispositions nor culturally evolved and transmitted regularities can be considered as the ultimate source of language universals. He draws a parallel with mathematical operations (addition, subtraction, etc.) and with prime numbers. Symbolic reference, he argues, is constrained by the structure it refers to. The editors claim, in the light of this diversity, that 'this book is intended to bring together, for the first time, all the major perspectives on language evolution'. We have two concerns with this aim. First, two books of the same organization and scope have been published in the past six years based on the material from language evolution conferences ([@pbio-0020346-Hurford1]; [@pbio-0020346-Knight1]). Although this first concern might be just splitting hairs, the second is more substantial: several crucial aspects of language evolution are not represented at all or are just touched superficially. One of these missing themes is the selective advantage of early language. As discussed, many of the contributors express their scepticism towards the selective scenarios found in the literature---and indeed towards such constructions in general---but there is no review and no balanced evaluation of these selective scenarios. Since one of the key questions of language evolution is the selective advantage of early language, the lack of such a review is a major weakness. A balanced account could have been presented even if the editors and most of the contributors are frustrated by the plethora of selective scenarios. Related to the possible selective advantage of language is the issue of genetic background. Although there is mention of the so-called FOX genes---some mutations of which are associated with language disorders---there is no detailed discussion of our current knowledge of genetics related to language. Another lightly treated theme is the neural basis of language and language evolution. Understandably it is one of the most difficult issues concerning human language, and no one expects the editors or any of the contributors to come up with an answer to all the questions. What is missing again is a good survey outlining the problems and the current findings of the field. The weaknesses of the book come from its structure and organization. The editors, instead of outlining a structure and asking specialists to contribute to that structure, appear to have let every contributor write freely about their current ideas and current research without regard to the bigger picture. This definitely shows the interests of the contributors and outlines the current state of the art; it leaves gaps, however, in the coverage of crucial topics related to the evolution of human language. Szabolcs Számadó and Eörs Szathmáry are at the Collegium Budapest (Institute for Advanced Study), Budapest, Hungary.	PubMed Central	2024-06-05T03:55:48.376361	2004-10-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521730/", "journal": "PLoS Biol. 2004 Oct 12; 2(10):e346", "authors": [ { "first": "Szabolcs", "last": "Számadó" }, { "first": "Eörs", "last": "Szathmáry" } ] }
PMC521731	""Some qualities nature carefully fixes and transmits, but some, and those the finer, she exhales with the breath of the individual as too costly to perpetuate. But I notice also that they may become fixed and permanent in any stock, by painting and repainting them on every individual, until at last nature adopts them and bakes them into her porcelain"---Ralph Waldo Emerson" The history of domesticated plant form and function evolves along a two-tiered track that doubles back on itself, offering panoramic vistas of natural forces intertwined with the creative force of human endeavor ([Figure 1](#pbio-0020347-g001){ref-type="fig"}). For approximately 10,000 years, human beings have modified the traits of plants and animals, giving rise to hundreds of thousands of domesticated breeds that today form the foundation of the world\'s food supply. Modern breeds are descendents of the wild species from which they were derived. The process of domestication dramatically changed the performance and genetic architecture of the ancestral species through the process of hybridization and selection as originally described by Charles [@pbio-0020347-Darwin1]. ::: {#pbio-0020347-g001 .fig} Figure 1 ::: {.caption} ###### The Diversity of Ancestral Rices \(A) Long, thin-grained rice with purple hull. \(B) Round-grained rice with white hull. \(C) Panicles of golden-hulled rice (foreground) and purple-leafed rice (background). \(D) Tall, weedy rice with pale leaves and silver hulls. ::: ![](pbio.0020347.g001) ::: Despite the low yields and poor eating quality of most wild ancestors and primitive crop varieties, these ancient sources of genetic variation continue to provide the basic building blocks from which all modern varieties are constructed. Breeders have discovered that genes hidden in these low-yielding ancestors can enhance the performance of some of the world\'s most productive crop varieties. In this essay, I will provide some historical context for the paper by Gur and Zamir in this issue of PLoS Biology ([@pbio-0020347-Gur1]). I will discuss how "smart breeding" recycles "old genes" to develop highly productive, stress-resistant modern varieties and why this approach is particularly attractive to increase food security in regions of the world with high concentrations of genetic diversity. The job of the plant breeder is to create an improved variety. This may be accomplished simply by selecting a superior individual from among a range of existing possibilities, or it may require that a breeder know how to efficiently swap or replace parts, recombine components, and rebuild a biological system that will be capable of growing vigorously and productively in the context of an agricultural environment. How the breeding is done and what goals are achieved is largely a matter of biological feasibility, consumer demand, and production economics. What is clear is that the surest way to succeed in a reasonable amount of time is to have access to a large and diverse pool of genetic variation. The process of plant breeding is theoretically simple, but its power resides in the fact that it creates novelty. A breeder generally selects two individuals for crossing, each of which has specific traits or characteristics of interest. The cross provides the mechanism by which genes are exchanged between the parents so that a wide array of diverse individuals is observed in the progeny of future generations. From a breeding perspective, this provides the basis for selection so that individuals containing the best features of both parents can be identified and further bred. By selecting parents that are genetically similar, a breeder restricts the amount of variation that will be evaluated in the offspring. On the other hand, by crossing genetically divergent parents, the range of phenotypic variation will be much more extensive and can even be surprising, with many individuals presenting phenotypes that would not be expected based on the attributes of the parents. Thus, if a breeder is interested in innovation and wants to generate maximum variation from which to make selections, wide crosses are the most productive. Not all genetic variation is created equal. When Darwin first introduced the concept of evolution ([@pbio-0020347-Darwin1]), he challenged the prevailing view that species were fixed entities with a single, invariable genetic identity. The concept of natural selection presupposed that species were comprised of genetically variable individuals such that selection could act on them. The genetic variants differ in the alleles (versions of genes) they carry. Alleles that are deleterious in terms of the survival and reproduction of the organism will eventually be eliminated while alleles that are favorable or neutral will be perpetuated in the population. Recombination in natural populations allows alleles that may be deleterious in one genetic background to be reassessed in a different genetic context. Over time, the alleles that are transmitted at high frequency across generations represent those with a substantial likelihood of contributing positively to an organism\'s long-term viability in a variable environment. For this reason, natural variation is a much more valuable and informative reservoir of genes for the purposes of plant improvement than would be an equivalent number of induced mutations generated in a laboratory. Domestication---The Winnowing of Natural Genetic Variation {#s2} ========================================================== Cultivars (domesticated varieties) have been selected by humans in the last 10,000 years and inevitably represent a subset of the variation found in their wild ancestors. Cultivars are recognizable because they manifest characteristics that are associated with domestication in plants. Unusual or extreme phenotypes, such as large fruit or seed size, intense color, sweet flavor, or pleasing aroma are often selected by humans and maintained in their cultivars for aesthetic reasons, while synchronous ripening or inhibition of seed shattering (a dispersal mechanism) are selected to facilitate harvest. These phenotypes may occur in nature but they will frequently be eliminated by natural selection before they are fixed in a population. Because of human selection, cultivars may exemplify a range of exaggerated phenotypic attributes that give them the appearance of being, on the whole, more diverse than some of the wild populations from which they were derived, but in truth, domestication usually represents a kind of genetic bottleneck. Furthermore, cultivars are grown in agricultural environments that are generally more uniform than the environments in which wild species grow, and this tends to further narrow the gene pool. Thus, while cultivars may embody a high degree of obvious phenotypic variation, this may not always be a good predictor of the extent of their genetic variation. The landrace varieties are the earliest form of cultivar and represent the first step in the domestication process. Landraces are highly heterogeneous, having been selected for subsistence agricultural environments where low, but stable yields were important and natural environmental fluctuation required a broad genetic base ([Figure 2](#pbio-0020347-g002){ref-type="fig"}). Landraces are closely related to the wild ancestors and embody a great deal more genetic variation than do modern, high-yielding varieties that are selected for optimal performance within a narrow range of highly managed environmental conditions. The value of both the wild species and the early landrace varieties in the context of modern plant breeding is that they provide a broad representation of the natural variation that is present in the species as a whole. The fact that natural selection has acted on such populations over the course of evolution makes them particularly valuable as materials for breeders. The value added by imposing a low intensity of human selection on the early landraces resides in the fact that some of these early varieties represent accumulations of alleles that produce phenotypes particularly favorable or attractive to the human eye, nose, palette, or other appetites. It is also noteworthy that some of these rare or unique alleles or allele combinations that were selected by humans might never survive in the wild. ::: {#pbio-0020347-g002 .fig} Figure 2 ::: {.caption} ###### Banaue Rice Terraces in the Philippines Where Traditional Landraces Have Been Grown for Thousands of Years ::: ![](pbio.0020347.g002) ::: Wild relatives and early landrace varieties have long been recognized as the essential pool of genetic variation that will drive the future of plant improvement ([@pbio-0020347-Bessey1]; [@pbio-0020347-Burbank1]). Early plant collections made by people such as Nikolai Vavilov (1887--1943) or Jack Harlan (1917--1998) inspired the international community to establish long-term collections of plant genetic resources that provide modern plant breeders with the material they need to creatively address the challenges of today ([Box 1](#box1){ref-type="boxed-text"}). Many may question the emphasis on wild and primitive landraces that cannot compete with new, high-yielding varieties in terms of productivity or eating quality, particularly in an age when biotechnology and genetic engineering promise to provide an endless stream of genetic novelty. Indeed, if all forms of novelty were equally valuable, the old varieties would hardly be worth saving. But the security of the world\'s food supply depends on an exquisite balance between new ideas and the intelligent use of time-tested resources. In 1972, more than a decade before the age of automated sequencing, Jack Harlan commented that, "We are not really much interested in conserving the old varieties as varieties; it is the genes we are concerned about. The old land races can be considered as populations of genes and genetic variability is absolutely essential for further improvement. In fact, variability is absolutely essential to even hold onto what we already have" ([@pbio-0020347-Harlan1]). Combining Breeding with Molecular Genetics {#s3} ========================================== In today\'s world where automated sequencing and DNA synthesis are mundane activities, it may seem contradictory to be worrying about saving or using "old genes." Can\'t new ones be synthesized to order? Can\'t we modify a plant at will by introducing a new gene or two into an existing variety? Why should we worry about saving populations of historically valuable genes in millions of living plant specimens at great cost to the tax-paying public? Perhaps it is not the genes themselves we are now in fear of losing. It is the information they encode in all their combinatorial complexity. After all, we are only at the very beginning of the endeavor to understand the way in which a genotype confers a particular set of attributes to a living organism. The subtleties of phenotypic plasticity in the face of a changing environment and the layers of genetic redundancy that characterize biological systems are largely mysterious. We have only just begun to consider the millions and billions of genetic trials and errors that have been evaluated by nature over evolutionary time. We cannot even begin to simulate the selective filters that have provided us with the diversity of form and function in the living world. We do know that living forms of natural diversity are needed to sustain life, and that it would be impossible to replace or recreate that diversity if it were lost at this time. As plant breeders, we know what to do with living forms of genetic diversity. If we keep our options open and learn to better utilize the reservoirs of natural variation that have been preserved in our gene banks and in the few remaining in situ populations of wild species and landrace varieties, an almost infinite array of novelty can be achieved using traditional, time-proven practices involving crossing and selection of genes that have withstood the test of evolutionary time ([@pbio-0020347-Burbank1]; [@pbio-0020347-Hawkes1]; [@pbio-0020347-Rick1]; [@pbio-0020347-Harlan3], [@pbio-0020347-Harlan4]; [@pbio-0020347-Peloquin1]). By restricting the gene pool, we can readily channel a phenotype into a constrained and predictable outcome. By expanding the gene pool, we can open up many new possibilities for consideration that have not been previously evaluated, would be unlikely to be generated in nature, and would not be readily predicted based on current knowledge. In crosses between wild and cultivated species of inbreeding plants, alleles that were "left behind" during the domestication process may be reintroduced into the cultivated gene pool. This infusion of "new blood" renews and invigorates modern cultivars in surprising and interesting ways. It is not uncommon for some of the inbred progenies derived from these crosses to perform better than the better parent ([@pbio-0020347-Frey1]; [@pbio-0020347-Rick3], [@pbio-0020347-Rick4]; [@pbio-0020347-Tanksley1]). This phenomenon is known as transgressive variation and results from positive interaction between the genotypes of the parents. Today, plant breeders can analyze populations derived from wide crosses using molecular markers to determine which portions of the chromosomes are associated with the transgressive variation of interest. This makes it possible to dissect a complex phenotype and to determine where individual genes or, more correctly, quantitative trait loci (QTLs) map along the chromosomes. Information about DNA markers linked to QTLs represents a powerful diagnostic tool that enables a breeder to select for specific introgressions of interest, a technique referred to as "marker-assisted selection." This approach has proven to be extremely successful in several crop species (tomato \[[@pbio-0020347-Bernacchi1]\], hybrid rice \[[@pbio-0020347-Xiao1]\], inbred rice \[[@pbio-0020347-Thomson1]\], wheat \[[@pbio-0020347-Huang1]\], barley \[[@pbio-0020347-Pillen1]\], and pepper \[[@pbio-0020347-Rao1]\]). In China, two introgressions from a wild relative of rice have been associated with a 30% increase in the yields of the world\'s highest-yielding hybrid rice ([@pbio-0020347-Deng1]). In tomato, yield increases of greater than 50% resulted from introgressing three independent segments from a wild relative, as reported by [@pbio-0020347-Gur1]. The effect of these introgressions on yield was stable in different genetic backgrounds and in both irrigated and drought conditions. This work was facilitated by the availability of a library of chromosome segment substitution lines, called introgression lines when the donor is a wild species, that provided the foundation for exploring the interactions among the independent QTLs. Plant geneticists have long recognized the value of exotic libraries (Brassica \[[@pbio-0020347-Ramsay1]; [@pbio-0020347-Cermankova1]\], millet \[[@pbio-0020347-Hash1]\], rice \[[@pbio-0020347-Sobrizal1]; [@pbio-0020347-Ghesquiere1]; [@pbio-0020347-Ahn1]\], tomato \[[@pbio-0020347-Monforte1]; [@pbio-0020347-Zamir1]\], wheat \[[@pbio-0020347-Sears1]; [@pbio-0020347-Pestsova1], [@pbio-0020347-Pestsova2]\], and Arabidopsis \[[@pbio-0020347-Koumproglou1]\]). They represent a permanent genetic resource that greatly facilitates the utilization of wild and exotic germplasm in a breeding program, and they are also an efficient reagent for the discovery and isolation of genes underlying traits of agricultural importance. Uncovering the Genes That Underlie Agronomic Traits {#s4} =================================================== Several genes underlying traits of agricultural importance have been cloned using substitution lines derived from interspecific or intersubspecific crosses ([@pbio-0020347-Martin1]; [@pbio-0020347-Song1]; [@pbio-0020347-Frary1]; [@pbio-0020347-Yano2]; [@pbio-0020347-Takahashi1]; [@pbio-0020347-Yano1]), including one of the yield QTLs targeted by Gur and Zamir ([@pbio-0020347-Fridman1]). While the identity of the yield gene conferring the phenotype was not critical to the success of the cultivar development scheme described by [@pbio-0020347-Gur1], there is great curiosity to understand the gene(s) or genes and genetic mechanisms that underlie traits of interest to agriculture. In some cases, knowing the gene or the exact functional nucleotide polymorphism within the gene that determines the phenotype ([@pbio-0020347-Bryan1]; [@pbio-0020347-Robin1]) may dramatically improve the resolution of selection during the breeding process. It also may allow a breeder to make more informed decisions about which germplasm resources to use as parents in a crossing program and which genes within those resources to use in a pyramiding scheme. As more genes of interest are cloned and their contributions to complex biological systems are better understood, there will be many opportunities for creative synthesis of new varieties. It is likely that some of the opportunities will involve genetic engineering approaches, where new information about genes, gene regulation, and plant responses to the environment may be used in innovative ways to fine-tune existing plant varieties so that they utilize resources more efficiently, provide greater nutritional value, or simply taste better. Natural Variation and Food Security {#s5} =================================== The scientific enterprise has always challenged beliefs about the way the world functions, its origins, and its possibilities. Deeply held beliefs are frequently resistant to the most carefully crafted scientific explanations. When belief systems are unconscious, they may prove particularly resilient to change. Occasionally, science provides an interpretation that fits cleanly into the framework of existing ideas, and then it is heralded with great applause, and often with a sense of relief. When this is not the case, public opinion tends to react fitfully, with many starts and stops. Public opinion has been on a roller coaster recently with respect to transgenic organisms in agriculture. This is in response to what is perceived to be a kind of scientific intrusion into the intimacy of the relationship between humans and their food supply. This relationship is inherently complex, representing a textured fabric of historical, cultural, geographic, economic, biological, and aesthetic concerns. Despite the fact that food is increasingly treated as a commodity in today\'s global economy, human culture the world over has always recognized that food represents more than a biological remedy for hunger. Food is a force that brings diverse people together, it provides a focal point for human discourse, and it enhances our enjoyment of life. Food also has a spiritual component. Harvesting other living organisms to support human life represents a powerful connection between different spheres of the natural world. At some level, the idea of using natural genetic variation found in wild species and early landrace varieties to revitalize modern crop varieties is both emotionally appealing and intellectually compelling. As a "smart breeding" strategy, it will facilitate the exploration and utilization of natural genetic variation, expanding the genetic base of our crop plants and providing more flexibility for the future. By using a marker-assisted approach, it will provide a noninvasive road map to expedite the selective introgression of useful traits in the years ahead. Because the approach is primarily useful for self-pollinating species (as opposed to cross-pollinators), variety development can go forward with the expectation that new varieties can be developed and distributed as inbred strains. This will come as very good news to people who are concerned about the infrastructural requirements needed to maintain a hybrid seed industry. Inbred variety seed can be saved from year to year without noticeable loss of vigor. Farmers are free to amplify the varieties and pass seed on to their neighbors if it proves valuable. Plant breeders living in parts of the world where germplasm diversity is highest are in the best position to explore its value. Until now, there have been few opportunities to make use of the wealth of natural diversity that abounds in many countries where people are the poorest and population is growing the fastest. This approach offers a way forward and can help people make good use of locally available resources to enhance the food security of their own nations. As we consider the implementation of smart breeding efforts in the future, we might ask, who will have access to nature\'s reserves of genetic diversity? How will knowledge about the patterns that govern the generation and selective elimination of that diversity help guide conservation efforts as well as current and future crop improvement efforts? What are the limits to biological variation? How far can we push those limits, and what will be the consequences of not pushing them? Who will participate in the endeavor? What will the rules of engagement be? What tools can we use to expedite the effort? What genetic characteristics will help us cope with climate change, global warming, the emergence of new pests and diseases, depleted soils, shortages of fresh water, and increasing levels of water and air pollution? What trace minerals, vitamins, and other metabolites will we need to breed into the crops of the future to fight the causes of hidden hunger, to prevent cancer, or to enhance the immune system? The combinatorial possibilities for crop improvement are almost infinite, as long as we maintain our options. Faced with a clear choice today, it is obvious that enhancing the potential for genetic flexibility in the future is a wise course of action and one we ignore at our peril. ::: {.caption} ###### Box 1. The Pioneers ::: "Moreover, from our wild plants, we may not only obtain new products but new vigor, new hardiness, new adaptive powers, and endless other desirable new qualities for our cultivated plants. All of these things are as immediate in possibilities and consequences as transcontinental railroads were fifty years ago."---[@pbio-0020347-Burbank1] Luther Burbank (1849--1926) was one of America\'s first and most prolific plant breeders. He was inspired by Charles Darwin\'s Variation of Animals and Plants under Domestication ([@pbio-0020347-Darwin2]) to explore the potential of creating new varieties of plants by cross-breeding (hybridization) and selection. Over a 50-year period, he developed more than 800 new varieties of fruits, vegetables, flowers, and grasses. One of his earliest creations was the Burbank potato (1871), a variety of baking potato still popular today. When the Plant Patent Act of 1930 was first introduced in Congress, Thomas Edison testified, "This \[bill\] will, I feel sure, give us many Burbanks." The bill passed, and Luther Burbank was awarded 16 posthumous patents for asexually reproduced plants ([@pbio-0020347-Burbank1]). Nikolai Vavilov (1887--1943), a Russian geneticist and biologist, was one of the first to explore and actively collect wild relatives and early landrace varieties as sources of genetic variation for the future of agriculture. His botanical collecting expeditions (1916--1940) amassed many thousands of rare and valuable specimens that are preserved in the Vavilov Institute of Plant Industry in St. Petersburg, the world\'s first seed bank and inspiration for the International Crop Germplasm Collections (<http://www.sgrp.cgiar.org/publications.html>). Vavilov\'s concepts in evolutionary genetics, such as the law of homologous series in variation ([@pbio-0020347-Vavilov1]) and the theory of centers of origin of cultivated plants ([@pbio-0020347-Vavilov2]), were major contributions to understanding the distribution of diversity around the world. Vavilov himself died of starvation in a Stalinist prison camp in 1943, victim of a debate about genetics at a time when Trofim Lysenko\'s theories about the alterability of organisms through directed environmental change proved more compelling to the Soviet leadership than Vavilov\'s own efforts to demonstrate the genetic value of wild and early landrace diversity. In the United States, Jack Harlan (1917--1998) was also well known for his plant collection expeditions and eloquent expositions about the value of wild relatives and early domesticated forms of crop plants ([@pbio-0020347-Harlan2]). What particularly sensitized Jack Harlan to the value of these genetic resources was the fact that he lived through a period of revolutionary change in the way agriculture was practiced, watching as the Green Revolution\'s high-yielding semi-dwarf varieties of wheat and rice replaced the old landrace varieties throughout Asia and Latin America ([@pbio-0020347-Harlan3]). He understood that the new varieties brought massive and immediate increases in grain production that saved millions from starvation. He also understood that displacement of the traditional varieties from their natural environment presented serious challenges that would require renewed efforts to collect, document, evaluate, and conserve plant genetic resources. "For the sake of future generations, we must collect and study wild and weedy relatives of our cultivated plants as well as the domesticated races. These resources stand between us and catastrophic starvation on a scale we cannot imagine" ([@pbio-0020347-Harlan2]). Charlie Rick (1915--2002) was an avid collector of exotic tomato germplasm. He noted that up until the 1940s, progress in tomato improvement lagged and few major innovations were achieved. The turning point, according to Rick, was the introduction of exotic germplasm. As a cultivated species, tomato had experienced a severe genetic bottleneck that led to extreme attrition of genetic variability compared to the wild species of Lycopersicon ([@pbio-0020347-Rick5]). Yet, Rick observed that crosses between wild and cultivated species generated a wide array of novel genetic variation in the offspring, despite the fact that routine evaluation of wild and exotic resources often failed to detect the genetic potential of these resources ([@pbio-0020347-Rick1], [@pbio-0020347-Rick2]). He outlined "pre-breeding" strategies that were designed to uncover positive transgressive variation in backcrossed (inbred) progeny derived from interspecific crosses and believed that this approach would invariably lead to greater utilization of the favorable attributes hidden in tomato exotics ([@pbio-0020347-Rick4]). Susan McCouch is in the Department of Plant Breeding and Genetics at Cornell University, Ithaca, New York, United States of America. E-mail: <srm4@cornell.edu> QTL : quantitative trait locus	PubMed Central	2024-06-05T03:55:48.377345	2004-10-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521731/", "journal": "PLoS Biol. 2004 Oct 12; 2(10):e347", "authors": [ { "first": "Susan", "last": "McCouch" } ] }
PMC521732	Skeletal muscle demonstrates a remarkable plasticity, adapting to a variety of external stimuli ([@pbio-0020348-Booth1]; [@pbio-0020348-Chibalin1]; [@pbio-0020348-Hawley1]; [@pbio-0020348-Fluck1]), including habitual level of contractile activity (e.g., endurance exercise training), loading state (e.g., resistance exercise training), substrate availability (e.g., macronutrient supply), and the prevailing environmental conditions (e.g., thermal stress). This phenomenon of plasticity is common to all vertebrates (Schiaffino and Reggiani 1996). However, there exists a large variation in the magnitude of adaptability among species, and between individuals within a species. Such variability partly explains the marked differences in aspects of physical performance, such as endurance or strength, between individuals, as well as the relationship of skeletal muscle fiber type composition to certain chronic disease states, including obesity and insulin resistance. In most mammals, skeletal muscle comprises about 55% of individual body mass and plays vital roles in locomotion, heat production during periods of cold stress, and overall metabolism ([Figure 1](#pbio-0020348-g001){ref-type="fig"}). Thus, knowledge of the molecular and cellular events that regulate skeletal muscle plasticity can define the potential for adaptation in performance and metabolism, as well as lead to the discovery of novel genes and pathways in common clinical disease states. ::: {#pbio-0020348-g001 .fig} Figure 1 ::: {.caption} ###### Anatomy of a Skeletal Muscle Individual bundles of muscle fibers are called fascicles. The cell membrane surrounding the muscle cell is the sarcolemma, and beneath the sarcolemma lies the sarcoplasm, which contains the cellular proteins, organelles, and myofibrils. The myofibrils are composed of two major types of protein filaments: the thinner actin filament, and the thicker myosin filament. The arrangement of these two protein filaments gives skeletal muscle its striated appearance. ::: ![](pbio.0020348.g001) ::: How Is Skeletal Muscle Fiber Type Classified? {#s2} ============================================= Much of our early understanding of the plasticity of skeletal muscle has been derived from studies undertaken by exercise physiologists (e.g., [@pbio-0020348-Holloszy1]). With the application of surgical techniques to exercise physiology in the late 1960s ([@pbio-0020348-Bergstrom1]), it became possible to obtain biopsy samples (∼150 mg) of human skeletal muscle, and by means of histological and biochemical analyses, specific morphological, contractile, and metabolic properties were identified. In 1873, the French anatomist Louis Antoine Ranvier had already observed that some muscles of the rabbit were redder in color, and contracted in a slower, more sustained manner, than paler muscles of the same animal. These early observations formed the basis of the classical terminology of red and white muscle fibers, which was subsequently found to be related to myoglobin (an iron-containing oxygen-transport protein in the red cells of the blood) content ([@pbio-0020348-Needham1]). Based upon histochemical staining ([@pbio-0020348-Engel1]), muscle fibers are now commonly distinguished as slow-twitch (ST), which stain dark or red, and fast-twitch (FT), which stain light or pale. In humans, a further subdivision of the FT fibers is made ([@pbio-0020348-Brooke1]), whereby the more aerobic (or oxidative) FT fiber is designated FT~a~, and the more anaerobic (glycolytic) fiber is termed FT~b~. Under aerobic conditions (sufficient oxygen supply to the working muscles), energy is produced without the production of lactate. Under anaerobic conditions (insufficient oxygen supply to the working muscles), energy is produced via the glycolytic pathway, which results in lactate accumulation and in turn limits anaerobic exercise. Thus, muscle fibers can be classified in terms of contractile and metabolic properties ([Table 1](#pbio-0020348-t001){ref-type="table"}). ::: {#pbio-0020348-t001 .table-wrap} Table 1 ::: {.caption} ###### Contractile Characteristics, Selected Enzyme Activities, and Morphological and Metabolic Properties of Human Skeletal Muscle Fiber Types ::: ![](pbio.0020348.t001) This table highlights the relationship between skeletal muscle fiber-type composition and the indicated contractile and metabolic properties thats are consistent with differences in speed and endurance. All values are expressed as a fold-change relative to ST oxidative fibers ::: All individuals have different capacities to perform aerobic or anaerobic exercise, partly depending on their muscle fiber composition. In untrained individuals, the proportion of ST fibers in the vastus lateralis muscle (the largest of the quadriceps muscles and the most commonly studied muscle in humans), is typically around 55%, with FT~a~ fibers being twice as common as FT~b~ fibers ([@pbio-0020348-Saltin1]). While marked differences in the metabolic potentials between FT~a~ and FT~b~ fibers are observed in untrained humans, the absolute level for the activities of oxidative and glycolytic enzymes in all fiber types is large enough to accommodate substantial aerobic and anaerobic metabolism ([@pbio-0020348-Saltin1]). While there is a large degree of homogeneity within individual skeletal muscles from rodents ([@pbio-0020348-Delp1]), this is not the case for humans ([@pbio-0020348-Saltin1]). The dramatic heterogeneity of fiber type composition between people may explain their remarkable variation in exercise performance. Does Muscle Fiber Type Composition Influence Athletic Performance? {#s3} ================================================================== During the 1970s and 1980s, it was popular to determine the muscle fiber composition of athletes from different sports events. These studies revealed that successful endurance athletes have relatively more ST than FT fibers in the trained musculature ([@pbio-0020348-Costill1]; [@pbio-0020348-Fink1]; [@pbio-0020348-Saltin1]). In contrast, sprinters have muscles that are composed predominantly of FT fibers ([@pbio-0020348-Costill1]). Accordingly, the belief that muscle fiber type can predict athletic success gained credibility. In particular, the notion that the proportion of ST fibers might be a factor governing success in endurance events was proposed ([@pbio-0020348-Gollnick1]; [@pbio-0020348-Costill1]). In this regard, the results of [@pbio-0020348-Fink1] are important. These researchers determined the fiber composition from the gastrocnemius muscle (the muscle of the calf of the leg) of 14 elite male long distance runners, 18 good (but not world-class) male long distance runners, and 19 untrained men. The elite group included Olympic medal winners ([Figure 2](#pbio-0020348-g002){ref-type="fig"}) and American record holders at the time. Muscle from the elite runners contained a larger proportion of ST fibers than either the good runners or the untrained men (79.0% ± 3.5% versus 61.8% ± 2.9% versus 57.7% ± 2.5% respectively; p \< 0.05). The values found for several of the elite runners were the highest observed in human muscle (\> 92% ST). Moreover, the ST fibers from the elite runners were 29% larger than FT fibers (p \< 0.05), and both ST and FT fibers were larger in the good runners than in the untrained men. Because of the marked hypertrophy (bulk increase) of the ST fibers in the elite runners, the cross-sectional area composed of these fibers was greater than either the good runners or the untrained subjects (82.9% ± 3.1% versus 62.1% ± 2.6% versus 60.0% ± 2.7% respectively; p \< 0.05). When the data from the elite and good runners was combined, a positive correlation between the proportion of ST fibers and the best 6-mile performance time was noted (r = −0.62, p \< 0.05). ::: {#pbio-0020348-g002 .fig} Figure 2 ::: {.caption} ###### Microscopic View of the Gastrocnemius Skeletal Muscle from a World-Class Marathon Runner, Frank Shorter (Olympic Gold Medalist, 1972; Olympic Silver Medalist, 1976) The darkly stained fibers are relatively slow in contractile rate and are ST. These fibers demonstrate a higher aerobic (oxidative) capacity and a lower anaerobic (glycolytic) potential than the lighter stained FT fibers. Shorter\'s muscle contains approximately 80% ST fibers. Reproduced with kind permission from David L. Costill and William J. Fink. ::: ![](pbio.0020348.g002) ::: However, fiber type alone did not determine the performances of the elite athletes. For example, two athletes with similar best times for the 42.2 km marathon distance (approximately 2 hr 18 min) had 50% versus 98% ST muscle fibers. Subsequent work ([@pbio-0020348-Foster1]) revealed that endurance running performance was better related to an athlete\'s maximal O~2~ uptake (VO~2max~; r= −0.84, −0.87, and −0.88 for 1-, 2-, and 6-mile times, respectively). Indeed, while an athlete\'s muscle fiber type is an important morphological component and is related to several contractile and metabolic properties (see [Table 1](#pbio-0020348-t001){ref-type="table"}), other physiological factors (e.g., VO~2max~, maximal cardiac output, and speed/power output at the lactate threshold) are more likely to determine the upper limits of endurance capacity ([@pbio-0020348-Coyle1]; [@pbio-0020348-Hawley2]). Do Alterations in Skeletal Muscle Fiber Type Contribute to Metabolic Disease? {#s4} ============================================================================= The close coupling between muscle fiber type and associated morphological, metabolic, and functional properties is not confined to athletic ability. Insulin sensitivity also correlates with the proportion of ST oxidative fibers ([@pbio-0020348-Lillioja1]). Specifically, insulin-stimulated glucose transport is greater in skeletal muscle enriched with ST muscle fibers ([@pbio-0020348-Henriksen1]; [@pbio-0020348-Song1]; [@pbio-0020348-Daugaard1]), thus priming ST muscle for accelerated glucose uptake and metabolism. A shift in fiber distribution from ST to FT fibers gives rise to altered activities of key oxidative and glycolytic enzymes ([@pbio-0020348-Pette1]). Indeed, the ratio between glycolytic and oxidative enzyme activities in the skeletal muscle of non-insulin-dependent diabetic or obese individuals is related to insulin resistance ([@pbio-0020348-Simoneau2]; [@pbio-0020348-Simoneau1]). Similarly, with ageing and physical inactivity, two other conditions associated with ST-toFT fiber-type transformation, oxidative capacity and insulin sensitivity, are diminished ([@pbio-0020348-Papa1]). Genes That Define Skeletal Muscle Phenotype {#s5} =========================================== Skeletal muscle fiber-type phenotype is regulated by several independent signaling pathways ([Figure 3](#pbio-0020348-g003){ref-type="fig"}). These include pathways involved with the Ras/mitogen-activated protein kinase (MAPK) ([@pbio-0020348-Murgia1]), calcineurin ([@pbio-0020348-Chin1]; [@pbio-0020348-Naya1]), calcium/calmodulin-dependent protein kinase IV ([@pbio-0020348-Wu2]), and the peroxisome proliferator γ coactivator 1 (PGC-1) ([@pbio-0020348-Lin1]). The Ras/MAPK signaling pathway links the motor neurons and signaling systems, coupling excitation and transcription regulation to promote the nerve-dependent induction of the slow program in regenerating muscle ([@pbio-0020348-Murgia1]). Calcineurin, a Ca^2+^/calmodulin-activated phosphatase implicated in nerve activity-dependent fiber-type specification in skeletal muscle, directly controls the phosphorylation state of the transcription factor NFAT, allowing for its translocation to the nucleus and leading to the activation of slow-type muscle proteins in cooperation with myocyte enhancer factor 2 (MEF2) proteins and other regulatory proteins ([@pbio-0020348-Chin1]; [@pbio-0020348-Serrano1]). Calcium-dependent Ca^2+^/calmodulin kinase activity is also upregulated by slow motor neuron activity, possibly because it amplifies the slow-type calcineurin-generated responses by promoting MEF2 transactivator functions and enhancing oxidative capacity through stimulation of mitochondrial biogenesis ([@pbio-0020348-Wu2]). ::: {#pbio-0020348-g003 .fig} Figure 3 ::: {.caption} ###### Exercise-Included Signaling Pathways in Skeletal Muscle That Determine Specialized Characteristics of ST and FT Muscle Fibers Contraction-induced changes in intracellular calcium or reactive oxygen species provide signals to diverse pathways that include the MAPKs, calcineurin and calcium/calmodulin-dependent protein kinase IV to activate transcription factors that regulate gene expression and enzyme activity in skeletal muscle. ::: ![](pbio.0020348.g003) ::: PGC1-α, a transcriptional coactivator of nuclear receptors important to the regulation of a number of mitochondrial genes involved in oxidative metabolism, directly interacts with MEF2 to synergistically activate selective ST muscle genes and also serves as a target for calcineurin signaling ([@pbio-0020348-Lin1]; [@pbio-0020348-Wu1]). New data presented in this issue of PLoS Biology ([@pbio-0020348-Wang1]) reveals that a peroxisome proliferator-activated receptor δ (PPARδ)-mediated transcriptional pathway is involved in the regulation of the skeletal musclefiber phenotype. Mice that harbor an activated form of PPARd display an "endurance" phenotype, with a coordinated increase in oxidative enzymes and mitochondrial biogenesis and an increased proportion of ST fibers. Thus---through functional genomics---calcineurin, calmodulin-dependent kinase, PGC-1α, and activated PPARδ form the basis of a signaling network that controls skeletal muscle fiber-type transformation and metabolic profiles that protect against insulin resistance and obesity. The transition from aerobic to anaerobic metabolism during intense work requires that several systems are rapidly activated to ensure a constant supply of ATP for the working muscles. These include a switch from fat-based to carbohydrate-based fuels, a redistribution of blood flow from nonworking to exercising muscles, and the removal of several of the byproducts of anaerobic metabolism, such as carbon dioxide and lactic acid. Some of these responses are governed by transcriptional control of the FT glycolytic phenotype. For example, skeletal muscle reprogramming from a ST glycolytic phenotype to a FT glycolytic phenotype involves the Six1/Eya1 complex, composed of members of the Six protein family ([@pbio-0020348-Grifone1]). Moreover, the Hypoxia Inducible Factor-1α (HIF-1α) has been identified as a master regulator for the expression of genes involved in essential hypoxic responses that maintain ATP levels in cells. In this issue of PLoS Biology ([@pbio-0020348-Mason1]), a key role for HIF-1α in mediating exercise-induced gene regulatory responses of glycolytic enzymes is revealed. Ablation of HIF-1α in skeletal muscle was associated with an increase in the activity of rate-limiting enzymes of the mitochondria, indicating that the citric acid cycle and increased fatty acid oxidation may be compensating for decreased flow through the glycolytic pathway in these animals. However, hypoxia-mediated HIF-1α responses are also linked to the regulation of mitochondrial dysfunction through the formation of excessive reactive oxygen species in mitochondria. Can You Become a Slow-Twitcher? {#s6} =============================== With the 2004 Olympics still fresh on our minds, many will ask: Who has the right stuff to go the distance? Athletes like Olympic champion Frank Shorter are clearly exceptional and represent an extreme in human skeletal muscle phenotype. Realistically, few of us can ever hope to run a marathon in world-class time. However, there may be cause for some optimism for the average mortal, since endurance exercise training in healthy humans leads to fiber-type specific increases in the abundance of PGC-1 and PPAR-α protein in skeletal muscle ([@pbio-0020348-Russell1]). Moreover, functional genomics support the concept that skeletal muscle remodeling to a ST phenotype, either through activated calcineurin or PPARδ, can protect against the development of dietary-induced insulin resistance ([@pbio-0020348-Ryder1]) and obesity ([@pbio-0020348-Wang1]). The results of these studies have clinical relevance since insulin-resistant elderly subjects and offspring of patients with type 2 diabetes mellitus have skeletal muscle mitochondrial dysfunction ([@pbio-0020348-Petersen1]; [@pbio-0020348-Petersen2]). Clearly, further translational studies in humans are required to test the hypothesis that increasing the proportion of ST oxidative muscle fibers will overcome the mitochondrial dysfunction and metabolic defects associated with insulin-resistant states. Juleen R. Zierath is with the Department of Surgical Sciences, Section of Integrative Physiology, Karolinska Institutet, in Stockholm, Sweden. John A. Hawley is with the Exercise Metabolism Group, School of Medical Sciences, Faculty of Life Sciences at RMIT University in Bundoora, Australia. FT : fast-twitch FT~a~ : aerobic FT fiber FT~b~ : anaerobic FT fiber HIF-1α : Hypoxia Inducible Factor-1α MAPK : mitogen-activated protein kinase MEF2 : myocyte enhancer factor 2 PGC-1 : peroxisome proliferator γ coactivator 1 PPARδ : peroxisome proliferator-activated receptor δ ST : slow-twitch VO~2~max : maximal O~2~ uptake	PubMed Central	2024-06-05T03:55:48.380663	2004-10-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521732/", "journal": "PLoS Biol. 2004 Oct 12; 2(10):e348", "authors": [ { "first": "Juleen R", "last": "Zierath" }, { "first": "John A", "last": "Hawley" } ] }
PMC521733	Most bird-pollinated flowers are both red and rich in nectar. The traditional explanation for this association is that, since red is inconspicuous to bees, it evolved to prevent bees from depleting the nectar of bird-pollinated flowers without effecting pollination. But bees can see, and they actually visit red flowers. So why are most bird-pollinated flowers red? To help answer this question, we need to consider how the outcomes of foraging decisions are affected by the community in which individuals live, and by the foraging options of other individuals. The Mystery {#s2} =========== Plants face a trade-off between attracting pollinators and remaining hidden from flower parasites (such as nectar robbers and seed predators). Consequently, there is often strong selection pressure for highly specific communication channels that can advertise the presence of their flowers to effective pollinators but not to other individuals. Many aspects of pollinator syndromes are best understood in these terms ([@pbio-0020350-Proctor1]). For example, flowers that are pollinated by birds---bird flowers---produce nectar at much higher rates than those pollinated by bees ([@pbio-0020350-Stiles1]). If a bee is attracted to such a flower, it might sometimes remove nectar and pollen without providing an outcrossing service (i.e., bringing pollen from a different plant of the same species) to the flower. Therefore, bird-pollinated flowers should advertise their presence to birds, but not to bees. Following this line of reasoning, Peter [@pbio-0020350-Raven1] suggested more than thirty years ago that bird-pollinated flowers were predominantly red because 'red is the only color of the spectrum that is at once inconspicuous to most insects and also an excellent "signal" of a high caloric reward for birds'. Raven\'s interpretation of inconspicuousness was soon transformed into invisibility; it was assumed that bees did not visit red flowers because they couldn\'t detect them ([@pbio-0020350-Proctor1]; Vogel 1996). However, this interpretation no longer holds. [@pbio-0020350-Chittka2] have shown that red flowers are not actually invisible to bees. Indeed, typical bird flowers with no UV reflectance, such as the scarlet gilia (Ipomopsis aggregata) and the scarlet monkeyflower (Mimulus cardinalis) ([Figure 1](#pbio-0020350-g001){ref-type="fig"}), are routinely visited and exploited by different bee species (reviewed by [@pbio-0020350-Chittka2]). Moreover, when bees are extremely abundant, they can drive birds away from red flowers. Echium wildpretii, an endemic of the Canary Islands, presents an entomophylous ('insect-loving') and an ornithophyllous ('bird-loving') subspecies ([Figure 2](#pbio-0020350-g002){ref-type="fig"}) that differ in flower colour: E. wildpretii trichosiphon, endemic to La Palma Island, has entomophylous, pink flowers, whereas E. wildpretii wildpretii, endemic to Tenerife Island, has ornithophyllous, red flowers, pollinated by generalist native birds and insects. E. wildpretii wildpretii is pollinated predominantly by birds early in the season until introduced honeybees (Apis mellifera, which have increased enormously in number because of apiculture) deplete the nectar and displace nectar-feeding birds ([@pbio-0020350-Valido1]). ::: {#pbio-0020350-g001 .fig} Figure 1 ::: {.caption} ###### Typical Bird-Pollinated Flowers \(A) Scarlet gilia Ipomopsis aggregata. Image courtesy of Clarence A. Rechenthin at US Department of Agriculture--National Resources Conservation Service PLANTS Database. (B) Scarlet monkeyflower, Mimulus cardinalis. Image by William & Wilma Follette at US Department of Agriculture--National Resources Conservation Service PLANTS Database ([@pbio-0020350-USDANRCS1]). ::: ![](pbio.0020350.g001) ::: ::: {#pbio-0020350-g002 .fig} Figure 2 ::: {.caption} ###### The Two Subspecies of Echium wildpretii \(A) The pink flowers of E. wildpretii trichosiphon are pollinated by insects. (B) The red flowers of E. wildpretii wildpretii are pollinated by generalist native birds, unless birds are driven away by large densities of bees. Photos courtesy of Alfredo Valido. ::: ![](pbio.0020350.g002) ::: So if red flowers are not invisible to bees, why are most bird-pollinated flowers red? Perhaps birds are particularly apt at detecting red objects ([@pbio-0020350-Chittka2])? Again, this is not strictly true. Although all birds detect red objects and some birds do have their greatest spectral sensitivity and finest hue discrimination towards the long-wavelength (red) end of the spectrum ([@pbio-0020350-Stiles1]), they can also respond to ultraviolet light, and there is no evidence that, for example, hummingbirds have greater spectral sensitivity or greater spectral discrimination ability in the red part of the spectrum ([@pbio-0020350-Goldsmith1]). Feeding experiments, where hummingbirds are given nectar in artificial flowers of different colours, show no inherited colour preferences; hummingbirds have temporary preferences that can be modified by conditioning ([@pbio-0020350-Proctor1]). So are there other clues as to how this mystery might be solved? The Visual System of Bees {#s3} ========================= One clue might come from the visual system of bees. Humans perceive light with a wavelength above approximately 600 nm as red ([@pbio-0020350-Buser1]). Most bees have three types of colour receptors, with sensitivity peaks at 340, 430, and 540 nm ([@pbio-0020350-Chittka1]), although a very few bee species have sensitivity peaks at substantially longer wavelengths. For the majority, however, provided that the light source is sufficiently intense, red light (up to 650 nm) will stimulate the 540 nm receptor of bees ([@pbio-0020350-Chittka2]). Bees will therefore perceive red objects. To discriminate red flowers from their green background, bees must rely essentially on the difference between the intensity of the signal that flower and foliage generate on the bees\' 'green' (540 nm) receptor ([@pbio-0020350-Giurfa1]). Therefore, depending on the relative intensity of the green and red sources, bees may or may not be able to discriminate between red flowers and green foliage ([@pbio-0020350-Chittka2]). Because of the structure of their visual system, bees trained to feed at artificial red flowers take longer to find their goals than bees trained to feed at other-coloured flowers ([@pbio-0020350-Spaethe1]). In a real environment, where red flowers would be more camouflaged against the different shades and intensities of the green foliage, the ability of bees to discriminate red flowers should be further reduced. Colour Vision and Niche Partition {#s4} ================================= The fact that bees require more time to find red flowers than other-coloured flowers, together with some results from optimal-foraging theory, outlined here, could unlock the mystery and explain the association between red coloration and bird pollination in flowers. When different animals, either from the same or different species, are forced to share some resources, any degree of specialization tends to result in habitat selection ([@pbio-0020350-Rosenzweig1]). In 1992, Possingham, developed a 'habitat selection' model that showed how two nectar-feeding pollinator species, which differed in their foraging efficiency, would forage on two types of flowers. Although an abstract model, we can use it to illustrate how birds might interact with bees at different-coloured flowers. Consider a community that includes bees and birds, and red and blue flowers. Let us assume that the flowers differ only in their colour, that there are only two patches of flowers (one of blue, the other of red flowers), and that the density of flowers is the same in both patches. (For a general analysis, with the same qualitative results, see [@pbio-0020350-Possingham1].) The question is: how many birds and bees should forage at the red and blue patches so that their intake of nectar is maximised? The expected intake rate is the average amount of nectar obtained per flower (or standing crop) divided by the time it takes to find and exploit a flower. If the flowers are the same distance apart and birds can detect red and blue flowers equally well, then travel time is independent of flower colour. Under these circumstances, an ecological equilibrium, with birds exploiting red and blue flowers equally, would indicate that the amount of nectar available from both flower colours was identical. Now add a few bees to this community of birds, sufficiently few that their intake of nectar is negligible. We know that the standing crop is the same at red and blue flowers. However, we also know that bees require more time to find red flowers than blue ones ([@pbio-0020350-Spaethe1]), so their intake rate of nectar will be higher at blue flowers, and they will all go to the blue patch. If we continue to add bees one at a time to this community, then sooner or later, the number of bees will no longer be sufficiently low for us to ignore their depleting effect on the nectar available. What will happen at that point? Will bees now start visiting red flowers? Not yet. For a bee to visit the red patch, the difference in standing crop between red and blue flowers would have to be large enough to compensate for the difference in detection time. Before that happens, some birds will shift to the red patch. Indeed, since birds require the same time to detect red or blue flowers, some birds will move from the blue to the red patch as soon as bees start to noticeably reduce the nectar available from the blue flowers. What Possingham\'s model predicts, therefore, is that when the number of bees is large enough, all birds will forage at the red patch. Only when the difference in standing crop between red and blue flowers is so large that it compensates for the reduced detectability of red flowers, will bees start visiting the red patch. To conclude, there will be an association between red flowers and birds. Birds will exploit red flowers, and bees blue flowers. In addition, depending on the relative abundance of bees and birds (and of red and blue flowers), either birds or bees, but never both simultaneously, can also exploit the other flower type ([Figure 3](#pbio-0020350-g003){ref-type="fig"}). ::: {#pbio-0020350-g003 .fig} Figure 3 ::: {.caption} ###### Possible Outcomes of [@pbio-0020350-Possingham1] Each panel represents one possible outcome. If red flowers predominate (left), bees forage at blue and red flowers, while all birds forage at red flowers. If blue and red flowers are equally abundant (middle), there is complete resource partitioning, with bees foraging at blue flowers and birds at red flowers. If blue flowers become more abundant (right), all bees forage at blue flowers, while birds forage at red and blue flowers. ::: ![](pbio.0020350.g003) ::: Niche Partition and the Evolution of Red Flowers {#s5} ================================================ [@pbio-0020350-Possingham1] helps to explain the ecological association between flower colour and pollinator type, provided that both flower colours and pollinator types are present---but why did the red colouration of these flowers evolve in the first place? We believe that the model can also help explain the evolution of red coloration in bird-pollinated flowers. To understand the evolutionary process, consider a community where bees and birds are present, and where two flower species coexist. One flower type, the generalist flower, is blue and is efficiently pollinated by bees and birds alike. The blue Rocky Mountain penstemon, Penstemon strictus, provides a good example ([@pbio-0020350-Castellanos1]). The other flower type, or bird flower, is yellow and is efficiently pollinated by birds, but not by bees---the red beardlip penstemon P. barbatus provides an example of this type ([@pbio-0020350-Castellanos1]). If bee visits were costly for the ancestral bird flowers, they would experience a selective pressure to become red. Bees could impose several costs on the ancestral bird flowers; for example, the number of hummingbird visits may depend on the amount of nectar available in the flowers. Throughout evolutionary history, there will be variability and heritability in flower colour (as documented for Mimulus by [@pbio-0020350-Bradshaw1]). Since both bees and birds easily detect and efficiently pollinate generalist blue flowers, there is no particular reason to expect that their colour will evolve in one direction or another. Things are otherwise for bird flowers, which are more efficiently pollinated by birds. For simplicity, consider that, at any given time, this bird flower comes in only two shades of colour, one of them with a slightly longer wavelength (an orange morph). On an ecological timescale, yellow flowers will be visited mainly by bees and orange flowers mainly by birds. Orange flowers, being more efficiently pollinated by birds, will therefore have higher fitness than yellow flowers, and given enough time, there will be selection for bird flowers to become orange. In the absence of other costs, mutant flowers with higher wavelengths (i.e., becoming redder) can invade a population of yellow flowers so long as bird flowers continue to be visited by bees (unpublished data). So bird flowers will continue to shift their colour until bees are completely excluded from the bird flowers or until further shifts deteriorate detectability by birds. This explanation for the evolution of red coloration in bird-pollinated flowers differs from the one proposed by [@pbio-0020350-Raven1] in a key respect. In our view, the main point is not that bees fly over red flowers without seeing them; it is not even that they are unable to exploit red flowers efficiently in absolute terms. It is rather a question of relative efficiency that makes bees avoid red flowers when birds are depleting their nectar; it would work just as well if birds were colourblind and perceived red flowers as badly as flowers of other colours. Of course, [@pbio-0020350-Possingham1] is not incompatible with birds being more efficient than bees at exploiting red flowers, and the results would be strengthened if, as has been suggested ([@pbio-0020350-Raven1]; [@pbio-0020350-Chittka2]), birds are better at detecting red flowers than blue ones. Toward a Solution {#s6} ================= Comparable problems can be found in other plant--pollinator systems. For example, when several species of bumblebees coexist, resource partitioning normally doesn\'t follow colour, but is dependent on different parameters: the corolla length of the plant and the proboscis length of the bee. Proboscis length affects the efficiency with which flowers of different depth are exploited ([@pbio-0020350-Inouye2]); bumblebees with long proboscises preferentially exploit flowers with deep corollas, while bumblebees with short proboscises exploit shallow flowers ([@pbio-0020350-Heinrich1]). But a bumblebee with a long proboscis can also exploit shallow flowers, and, to some extent, a bumblebee with a short proboscis can exploit deep flowers, if corollas are not too deep (although they will still leave some nectar behind). Indeed, when one bumblebee type is experimentally removed, the other one is seen to exploit both deep and shallow flowers ([@pbio-0020350-Inouye1]). The same, we believe, should happen with flower colour: the experimental removal of birds should lead to the systematic exploitation of red flowers by bees, at least when corolla tube morphology does not prevent bees from accessing the nectar. In fact, there is even no need to perform experimental bird removals, because plants provide us with a ready-made design: bees visit flowers searching for both nectar and pollen, while most birds exploit only the nectar. Hence, bees should readily collect pollen at red bird flowers. There are numerous examples of this, although in most cases they are indirectly documented. For example, solitary bees and syrphid and muscoid flies visit the red, hummingbird-pollinated flowers of Ipomopsis aggregata to collect pollen when hummingbirds visits are frequent, while bumblebee (Bombus appositus) visits to collect nectar are only common when hummingbird visits are rare ([@pbio-0020350-Mayfield1]). Outside the native range of bird-pollinated plants, the same phenomenon can be observed: in Spanish gardens, the honeybee collects pollen from Aloe arborescens plants. Bees cannot access the nectar, concealed at the bottom of the corolla tube. This is opportunistically collected by birds such as the Sardinian warbler Sylvia melanocephala (unpublished data). Another comparison of interest concerns beetle-pollinated flowers, which in the Mediterranean region have open, bowl shapes and red coloration ([@pbio-0020350-Dafni1]). Amphicoma beetles are more efficient pollinators of these flowers than commonly occurring bees ([@pbio-0020350-Dafni1]), so the red coloration of these flowers might help to keep other visitors (possibly bees and flies) at bay. Indeed, other bowl-shaped flowers of different colours (such as yellow, white, and purple, e.g., in the genera Cistus and Helianthemum) are commonly visited by pollen-collecting bees and bumblebees. A particularly interesting test case is provided by the corn poppy Papaver rhoeas; in the eastern Mediterranean region, it is pollinated by beetles and does not reflect in the UV ([@pbio-0020350-Dafni1]), while in central and western Europe it reflects in the UV ([@pbio-0020350-Daumer1]) and is pollinated by bees. Although refinements of Possingham\'s model, such as developing a prey-model version, or introducing stochasticity or several foraging constraints, might help us determine the extent to which we should expect resource partitioning along the colour dimension to take place, it is, in our view, far more pressing to determine the extent and conditions under which bees exploit red flowers (i.e., through comparisons of pollen vs. nectar exploitation, bird exclusion experiments, etc.), the detection time of red flowers against a natural background, and the effect of flower colour and size on flight mode in the field. Only then will we be able to fully unravel the factors that solve this fascinating mystery. We thank A. Traveset, C. Herrera, and an anonymous referee for their critical comments on former versions of the manuscript. Part of this work was conducted during MARG\'s fellowship at the Institute for Advanced Study of Berlin. The work was supported by funding from the Spanish Ministry of Science and Technology (INVASRED project, Ref. REN2003-06962/GLO). Miguel A. Rodríguez-Gironés is with the Estación Experimental de Zonas Áridas, Consejo Superior de Investigaciones Científicas in Almería, Spain. Luis Santamaría is with the Mediterranean Institute for Advanced Studies, Consejo Superior de Investigaciones Científicas in Mallorca, Spain.	PubMed Central	2024-06-05T03:55:48.382200	2004-10-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521733/", "journal": "PLoS Biol. 2004 Oct 12; 2(10):e350", "authors": [ { "first": "Miguel A", "last": "Rodríguez-Gironés" }, { "first": "Luis", "last": "Santamaría" } ] }
PMC521734	*In PLoS Biology, volume 2, issue 8* Nicotine\'s Defensive Function in Nature Anke Steppuhn, Klaus Gase, Bernd Krock, Rayko Halitschke, Ian T. Baldwin DOI: [10.1371/journal.pbio.0020217](10.1371/journal.pbio.0020217) The Academic Editor was erroneously listed as Michael Levine. The Academic Editor for this paper is Joy Bergelson, Chicago University. This correction note may be found online at DOI: 10.1371/journal.pbio.0020382.	PubMed Central	2024-06-05T03:55:48.384114	2004-10-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC521734/", "journal": "PLoS Biol. 2004 Oct 12; 2(10):e382", "authors": [] }
PMC522747	Background ========== The \'genomics revolution\' has provided the information needed to analyze how a genome responds to the environment in the formation of the \"transcriptome\", the portion of the genome that is transcribed. Microarrays, which offer the ability to analyze the expression ratios (ERs) of thousands of genes simultaneously, represent one of many new tools produced by this effort. However, not all biological disciplines have benefited equally from this technology, and array technology has not been widely adopted by the ecological community for a number of reasons. The large genome-wide arrays are only available for select model organisms, which may not be appropriate for many ecological questions. Moreover, the complexity of their analysis and the costs of the available commercial software solutions prohibit their adoption by small research groups and constrain the number of biological experiments that can be conducted even by large, better-funded groups. \'Boutique\' arrays -- on which a smaller fraction of the transcriptome, typically representing a selection (hundreds) of genes specific to a class of genetic elements or response types -- are readily created for a non-model organism at costs that are affordable for small research groups. However, the problems remain of how best to normalize and analyze array data. A large number of software solutions are available but no clear best solution has emerged \[[@B1]-[@B5]\]. A recent review has examined the types of arrays as well as the ecological and evolutionary questions that can be addressed with microarrays \[[@B6]\]. Here we present a cDNA microarray designed to analyze plant-herbivore interactions in a native plant. A cDNA microarray is a comparative tool, providing relative ERs for multiple genes from two differentially labeled fluorescent cDNA samples prepared by reverse transcription of mRNA extracted from matched plant samples. Hence the procedure is particularly useful for the analysis of plant responses elicited by herbivore attack: the induced defense and tolerance responses of plants \[[@B7]\]. We examine a number of practical challenges facing the adoption of boutique arrays for ecological research with tools familiar to ecologists, including signal normalization, the use of arbitrary expression thresholds to determine the significance of expression, the use of within-and between-array signal variance in evaluating the effect of probe quality and quantity and array age, as well as data analysis and visualization by cluster and Principal Component Analysis (PCA). The microarray was designed to examine herbivore-induced gene expression in the model ecological expression system, Nicotiana attenuata\[[@B8]\]. The genes for the microarrays were derived from a series of display (differential display reverse transcriptase-PCR, subtractive hybridization with magnetic beads, and cDNA-AFLP display) experiments that compared the transcriptome of plants attacked by the larvae of its specialist herbivore, Manduca sexta, with that of unattacked plants \[[@B9]-[@B11]\]. Two independent and differentially end-labeled cDNA probes of each of 240 genes were spotted in quadruplicate on each array. Hence each gene was represented by 8 replicate probes, which were used to analyze within-array ER variance (array CV). Since the array was composed of genes that were both down-or up-regulated in response to M. sextaattack, an array-specific normalization factor could be readily calculated. The effects of microarray age and cDNA quality on the measures of array CV were estimated. We present a 2-step criterion for determining significant expression based on t-tests of replicate ERs and arbitrary thresholds. We re-examine the use of arbitrary expression thresholds with a \'norm of reaction\' analysis of 6 genes derived from the 73 hybridization experiments. The data from microarrays are frequently analyzed with cluster analysis procedures \[[@B12]\], which deliver a limited analysis of the statistical associations. PCA is frequently used in ecological studies but is not commonly used in the analysis of array data. We present a PCA of 35 hybridized arrays, which visualizes the contribution of ERs from particular genes to the associations among arrays in the PCA. Results and Discussion ====================== Array CVs, array age and probe quality -------------------------------------- The array CV for each of the 73 arrays was strongly correlated with the number of gene ERs that showed higher values than the defined threshold for the variance (R^2^= 0.969, F~69,1~= 2102, P\< 0.001). This demonstrates that the array CV corresponds to the number of gene ERs that are outliers and therefore reflects the \"quality\" of the information derived from the array. We used array CVs to test if array age could explain some of the variance and found no detectable effect (R^2^= 0.025, F~69,1~= 2.73, P= 0.103). The spectrum of the cDNA was recorded between 240 and 700 nm. Shape and maxima of the curves for the particular compounds (DNA, Cy3, Cy5) allowed the evaluation of cDNA quantity and quality. The quantity of the cDNA that was hybridized was estimated by its OD at 260 nm. The quality of the fluorescently labeled probe derived from this cDNA was estimated by the relation of the quantity of the two dyes at 550 nm (Cy3) and 650 nm (Cy5) and the cDNA quantity. These linear regressions revealed that for cDNA quantity (OD 260 values), neither Cy3 nor Cy5 values were significantly correlated with array CVs (all R^2^\< 0.007, all Ps \> 0.225). There was a negative correlation between array CV and OD 550 values for Cy3 (R^2^= 0.069) and OD 650 values for Cy5 (R^2^= 0.051) with slopes of -7.55 and -5.7, respectively. However, only the Cy3 regression was significant (P = 0.028) whereas Cy5 was not (P = 0.06). A similar pattern was apparent for the probe quality: Cy3 and Cy5 quality parameters were negatively correlated with array CV, but only the regression for Cy3 (R^2^= 0.144, slope = -0.31) was significant (P = 0.001) whereas the regression for Cy5 (R^2^= 0.042, slope = -0.126) was not (P = 0.088). From this analysis, we conclude that the quality of the labeled cDNA sample to be hybridized to an array predicts the quality of the signals produced from the array. PCR product quality ------------------- The 502 different PCR products (2 for each gene + internal controls) that were spotted on the arrays had the following distribution in the 4 quality categories (Fig [2A](#F2){ref-type="fig"}): 1 = single band (426): 2 = single band with slight background (48); 3 = single band with strong background (14); 4 = multiple bands with background or only background (14). Multiple bands were only spotted to determine how low quality PCR products effect the results. To evaluate the association of PCR product quality on the variance of ERs, gene CVs were plotted against the PCR quality class. Gene CVs were found to be significantly different among the 4 PCR categories (Fig. [2B](#F2){ref-type="fig"}, Kruskal-Wallis ANOVA on Ranks, H~3~= 40.603, P\< 0.001). While post hoctests revealed that PCR quality did not have a directional effect on gene CV, it was lowest for genes with intermediate CVs and increased in genes with high and low CVs. We conclude that the quality of the PCR product spotted on arrays does not have a strong effect on gene CV. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### A: 2% agarose gel with 1 kb and 100 bp size ladders and examples illustrating the 4 different PCR-qualities classes (1 -- 4): 1 = single band: 2 = single band with slight background indicating multiple non-specific PCR products; 3 = single band with strong background; 4 = multiple bands with background or only background. B:Mean coefficient of variance (CV) of expression ratios for 8 replicate cDNA products from 253 genes (array CV) measured from73 hybridized microarrays based on the 4 PCR quality classes described in A. Classes have significantly different CVs (Kruskal-Wallis ANOVA on Ranks, H~3~= 40.603, P\< 0.001) ::: ![](1472-6785-4-13-2) ::: Expression patterns ------------------- ### All arrays A cluster analysis of 35 arrays (Fig. [3](#F3){ref-type="fig"}) reflected the similarities of the transcriptional patterns observed in arrays hybridized with similar treatments. Arrays hybridized with probes derived from mRNA from N. clevelandii(arrays 12, 13, 14) and N. longiflora(arrays 17, 18, 19, 32) were separated from those hybridized with material from N. quadrivalis(arrays 10, 11) and N. attenuatathat had been attacked by aphids or leaf hoppers (arrays 15, 16, 26, 27). These arrays were separated from those hybridized with samples from antisense-transformed N. attenuataplants that had been attacked by Manducacaterpillars (arrays 25, 28, 29, 30, 31, 34, 35), and the cluster they formed was separated from all other arrays that had been hybridized with N. attenuatamaterial elicited by methyl jasmonate treatments (MeJA, arrays 9, 20, 21, 22, 23, 24) or attacked by Manduca, mirid, Heliothisor Spodopteraherbivores (arrays 1 -- 8). The 3 replicated arrays hybridized with the same mRNA clustered together (arrays 33, 34, 35). The details of these similarities will not be treated here, as they are discussed in separate publications. The similarity of the elicited transcriptional signatures observed on the arrays hybridized with the N. longifloraand N. attenuata\[[@B13]\] probes demonstrates the utility of the array in the analysis of samples from congenerics. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Cluster analysis (Ward\'s method, squared Euclidean distance) showing similarities between 35 hybridized microarrays hybridized with probes from wildtype (WT), antisense Lox-3 (AS LOX) and antisense TD (AS TD) of the diploid native tobacco species, Nicotiana attenuataplants and from untransformed plants of 3 congeneric Nicotianaspecies, two of which are allotetraploids N. quadrivalisand N. clevelandiiand that are thought to have N. attenuataas a common ancestor as well as the more distantly related, N. longiflora. Arrays were hybridized with probes from plants attacked by different herbivore species or elicited with methyl jasmonate (MeJA). The shaded box represents 3 replicate hybridizations of the same sample of m-RNA from LOX N. attenuataplants. Arrays included in brackets correspond to clusters in the PCA (Fig. 4). ::: ![](1472-6785-4-13-3) ::: A PCA of the same 35 arrays (Fig. [4](#F4){ref-type="fig"}) showed a similar pattern of associations but provided the additional information of which genes contributed most to the patterns observed in the PCA. The vector of the gene coding for proteinase inhibitors (PI) was correlated with the first canonical axis that explained 40% of the total variance in the dataset. Moreover, transcripts for PI were up-regulated in the N. attenuataarrays elicited with MeJA or attacked by Manduca, mirid, Heliothis or Spodopteraherbivores. Expression of xyloglucan endo-transglycosylases (XTH) and WRKYtranscription factor transcripts was also correlated with the first axis and correlated with the location of arrays 1 -- 8 in the PCA. These 2 genes were plotted relatively close together in the PCA, reflecting their similar patterns of regulation across all arrays. The vector of allene oxide synthase (AOS) transcript expression reveals a correlation with arrays hybridized with mRNA from MeJA-elicited plants. AOScatalyzes a later stage in the biosynthesis of jasmonic acid and is known to be elicited by MeJA treatments \[[@B14]\]. The response of two unknown genes (434and 540) exemplifies genes whose pattern of expression is opposite to that discussed for the genes of known function. The ERs of gene 434reacted in the opposite direction as those of WRKYand XTH, and the responses were larger in antisense N. attenuataplants. The response of gene 540was opposite to that of AOSand larger in N. clevelandiiand N. attenuataplants attacked by leaf hoppers. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Principal component analysis (PCA) of the distribution of mean gene expression ratios of 234 genes (origin of vector is at the intersection of Axes 1 and 2 and the top of vector plotted as triangles) in the 35 hybridized microarrays (plotted as circles and squares) hybridized as described in Fig. 3. Arrays are labeled according to plant species and treatment (see Fig 3). Particular genes are identified with arrows. Clusters of microarrays with similar treatments are connected by lines. Axes 1 and 2 account for 40.1 and 8.3 % of the variation, respectively. Vectors of genes that are relatively long and parallel and, as such, correlated with the first canonical axis explain a large part of the variance. Vectors are strongly (up-or down) regulated in arrays and clusters of arrays lying near the end of a particular gene vector. For example, the WRKYgene vectors (originating at the intersection of Axes 1 and 2 and terminating at the WRKYgene triangle) contribute significantly to the cluster of microarrays hybridized with labeled cDNA derived from MeJA-elicited and Manducaand mirid attacked plants. Or, for example, the AOSgene vectors (originating at the intersection of Axes 1 and 2 and terminating at the AOSgene triangle) contribute significantly to the clusters of microarrays hybridized with labeled cDNA from MeJA-treated plant. ::: ![](1472-6785-4-13-4) ::: ### Individual genes The expression patterns of 6 genes (4 of known function; 2 of unknown function), as the mean of 2 PCR fragments with differently modified primers across 73 experiments, illustrate the \'norm of reactions\' of the transcriptional responses of these genes (Fig. [1](#F1){ref-type="fig"}). The transcriptional responses of these genes were in opposite directions and within various ranges of expression to the different treatments. Genes such as PIand XTHexhibit strong up-regulation (up to 88-fold) in response to herbivore attack and jasmonate elicitation, and are similarly strongly down-regulated (50-fold) when plants transformed to silence endogenous jasmonate biosynthetic enzymes (antisenseLOX\[[@B15]\]) are elicited and compared with untransformed plants on the same array. The inset of the XTHnorm of reaction depicts the variance in ERs from a selection of individual arrays to illustrate that treatments (arrays 16 and 6) eliciting very similar mean ERs (both 3.20) can have very different gene standard errors of the mean ER (SE) (0.19 and 0.78 respectively). Shaded areas represent the arbitrary ER thresholds of ± 0.3 for log~2~-transformed values. All 6 genes had numerous treatments in which these thresholds were exceeded, but the genes differed in the magnitude and direction in which the threshold ERs were exceeded. In contrast to the PIand XTHgenes, the unknown gene 540and the WRKYtranscription factor had more attenuated \'norm of reactions\', being maximally up-and down-regulated by only 6.5-and 5-fold across all 73 arrays. In a majority of the experiments, the AOSgene was up-regulated, while down-regulation was more common for unknown gene 540. In many experiments, however, ERs did not exceed threshold values. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Norm of reaction of expression ratios (ER) for 6 genes from 73 hybridized microarrays of which the 35 presented in the cluster analysis (Fig. 3) are labeled with numbers. Distribution of log-transformed mean expression ratios of 4 genes of known function \[proteinase inhibitor (PI), xyloglucan endo-transglycosylase (XTH), NtWRKY2 (WRKY) and allene oxide synthase (AOS)\] and 2 of unknown function (540and 434); dotted areas represent the arbitrary ER thresholds of 1.24 and 0.81 (corresponding to ± 0.3 for log~2~transformed values). Genes are organized according to the relative spread of their expression ratios A \> B \> C. Insert in XTHpanel shows error structure (mean ± SE) on a non-log scale calculated from 8 replicate spots from each array. ::: ![](1472-6785-4-13-1) ::: Ecologists are frequently interested in the processes that \"fit\" organisms to their environment. Adaptation to a particular environment results in part from the phenotypic consequences of hundreds of coordinated changes in gene expression, but because many levels of organization exist between an organism\'s transcriptome and its phenotype, it is often unclear how best to study the process of adaptation. Array technology has the potential to identify genes relevant to the process of adaptation, regardless of the time scale involved (evolutionary to physiological). However, a number of technical issues remain to be solved before the technology can be fully incorporated into ecological research: the normalization of signals, the within-and between-array variability of ERs, and the general problem of coping with the large amount of data that array studies produce. Many techniques have been discussed but a consensus for a standard solution \[[@B4]\] has not yet emerged. ### Normalization Since mRNA samples are labeled with different efficiencies and the different fluorescent dyes have different optical properties, signals from an array require normalization before ERs can be calculated. The literature addressing the problems of normalization has been reviewed \[[@B1],[@B2]\], with the consensus conclusion that there is no single best way to normalize array data and that specific solutions were required for the particularities of each array. When arrays are created with cDNAs that are typically both up-and down-regulated, a total intensity normalization can be used. By adapting a total intensity procedure \[[@B1]\] we normalized the signals from only the middle 75% of the distribution from a given array which produced values that were highly comparable among arrays, as demonstrated by the similarity of the clustering of the 3 replicate arrays (arrays 33, 34, 35; Fig [3](#F3){ref-type="fig"}, Fig. [4](#F4){ref-type="fig"}). ### Variance ERs from microarrays are derived from two differently labeled but mixed samples that competitively hybridize to immobilized gene-specific probes. The outcome of this hybridization can vary substantially within an array, as measured by the variance in ERs measured across replicate spots. The strong positive correlation between the number of genes above the specified ER-threshold and the array CVs highlights the utility of array CVs to summarize the quality of a given hybridization. Little is known about the factors that influence within-array hybridization or the amount of spot replication that is required to cope with the variance typically found in environmental samples \[[@B16]\]. However, the 8 replicate spots for each gene distributed across the array provided valuable data on gene and array CVs. From these CVs we were able to determine the quality of ER patterns from single arrays and single genes. Most of the technical parameters tested were not correlated with the variance structure in our dataset. Our measures of PCR product quality did not explain the variance of gene CVs. Similarly, array age did not account for a significant amount of variation in array CVs. In contrast, probe quality was negatively correlated with array CV and explains a part (ANOVA, F~69,1~= 5.046, P= 0.028) of the variance in array CVs. In our data set, a 15-fold increase in OD was associated with a halving of array CV. Therefore the monitoring of this measure of probe quality could save the costly use of arrays for samples that will likely produce low-quality results. Since none of the measured parameters unambiguously explained the pattern of within-array variance in our dataset, we conclude that a combination of several factors including the probe quality determines array variance. ### Data analysis Cluster analysis revealed groups of treatment that resulted in similar patterns of expression and, in doing so, provided a visual demonstration that the results obtained were reproducible. The PCA proved to be more useful for exploratory data analysis than did cluster analysis, because it provided information on the single gene vectors that contributed to similarities and differences among arrays. PCA allows researchers to quickly visualize similarities in expression patterns between known and unknown genes, and thereby generates hypotheses about the function and regulation of genes of unknown function. For example, in our analysis, a group of unknown genes -- from which we chose two (434 and 540) as proxies -- explained a relatively large part of the variance (indicated by long vectors) and was positively correlated to specific treatments and negatively correlated to vectors of genes of known function. Gene 434was up-regulated in antisense LOXplants and had the opposite pattern of expression compared to that of the WRKYand XTHgenes, both of which are strongly up-regulated by herbivore attack and jasmonate elicitation. Gene 540had the opposite pattern of regulation as did AOSwith higher ERs in plants attacked by leaf hoppers, suggesting a role in the plants\' response to this herbivore. The PCA of Fig. [4](#F4){ref-type="fig"} is a 2-dimensional presentation of a multidimensional analysis and analyses that allow for multidimensional presentations of the associations, provide more accurate information on the contribution of single gene vectors to associations among arrays. Quantitative geneticists have coined the term \'norm of reaction\' for the variation in phenotypic expression of a given genotype across a number of different environments. We apply this term to characterize the range of ERs observed for a given gene across a number of different expression experiments. The information provided in a norm of reaction provides a biologically informed alternative to the use of arbitrary thresholds for the determination of significant expression. This would allow researchers to use lower thresholds for genes (e.g. WRKYtranscriptions factors) that are known to show low dynamic ranges of expression and higher thresholds for genes with likely larger dynamic ranges, such as those directly involved in defense metabolite production (e.g. PI). Additionally, when comparing many arrays, a norm of reactions provides information that allows researchers to determine if a given array is providing ERs within the normal range of variance found in prior experiments. Conclusions =========== We conclude that the data produced by \'boutique\' microarrays can be readily analyzed with inexpensive home-grown procedures that are commonly used in ecological studies. Arrays with sufficient within-array replication allow for the calculation of gene and array CVs that are useful in estimating the quality of the information gathered from a given array. Furthermore, multivariate statistical techniques, such as PCA, can be used to visualize global expression patterns and identify the individual genes that make large contributions to the transcriptional signatures of particular treatments. The costs of boutique arrays are approaching those of many standard ecological procedures, and the information they provide will allow ecological researchers the ability to characterize early stages in an organisms\' response to environmental changes. Methods ======= Microarray construction and hybridization ----------------------------------------- The cDNA microarray and its hybridization is described in \[[@B11]\], and a complete list of cDNAs and their physical location on the microarray can be found at: \[[@B11]\] (supplemental Table I at <http://www.plantphysiol.org/content/vol131/issue4/images/data/1894/DC1/webpage-table1.xls>). Briefly, the production of the cDNA microarray started with a set of 234 genes which were cloned by differential display reverse transcription (DDRT)-PCR and subtractive hybridization using magnetic beads (SHMB) of M. sextalarvae-attacked N. attenuataplants \[[@B9],[@B10]\] or by cDNA-AFLP (amplified fragment-length polymorphism) display of N. attenuataplants under simulated M. sextaattack by applying oral secretions and regurgitant to leaf wounds \[[@B11]\] and 6 well-characterized Manduca-induced genes (putrescine methyl transferase, xyloglucan-8 endotransglycosylase, allene-oxide synthase, hydroperoxide lyase, trypsin inhibitor, WRKYtranscription factor). These genes were PCR amplified and for each cDNA, two PCR fragments, with 5\'-aminolink on either strand, were synthesized. Each PCR fragment was robotically spotted four times on epoxy coated slides (Quantifoil Micro Tools GmbH, Jena); hence, each gene was represented on the microarray 8 times: by two independent PCR fragments, which, in turn, were each spotted in quadruplicate. The cDNA microarrays were hybridized with fluorescently labeled cDNA prepared by reverse transcription of mRNA isolated from leaf tissues of 73 differently elicited Nicotianaplants belonging to 4 species. Competitive hybridization of 2 samples (treated and untreated plants) with different dyes (Cy3 and Cy5) defined the ratio of transcript abundance in the treatment sample compared to the control sample for each spot on the microarray. A majority of the arrays were hybridized with samples from wildtype or transformed \[[@B17]\]N. attenuataplants, which were elicited by attack from either various herbivore species (larvae of Manduca, Heliothis, Spodopteramoths, and adults and nymphs of aphids and mirids that attack N. attenuata), methyl jasmonate (MeJA), or larval regurgitant treatments or UV-B exposure, and compared with plants of the same genotype, age, and developmental stage which were unelicited. To determine the utility of the arrays in the analysis of responses from congenerics, arrays were hybridized with samples taken from two tetraploid species (N. quadrivalisand N. clevelandii) that had evolved from independent allopolyploid hybridizations between N. attenuataand another extinct 12-chromosome Nicotianataxa \[[@B18]\], as well as the more distantly related, N. longiflora. The details of each hybridization and the specific gene responses of the arrays are either published \[[@B11],[@B13],[@B19]\] or are in preparation. Here we present a global analysis of 73 arrays to identify methods of analysis for such boutique microarrays that are useful for ecological research. Normalization and statistics ---------------------------- Because the arrays included both up- and down-regulated genes, the calculation of a microarray-specific normalization factor provided a valuable alternative to the use of external reference controls, which may or may not be influenced by the elicitation conditions \[[@B2],[@B20]-[@B22]\]. The measured Cy5 and Cy3 fluorescence intensities were ranked independently, and after discarding the 12.5% maximum and minimum values, the remaining 75% of the values were summed (adapted total intensity normalization, \[[@B1]\]). The array-specific normalization factor was obtained by dividing the calculated sum of Cy3 values by those of the Cy5 values. The ratios of normalized fluorescence values for Cy3 and Cy5 of each individual spot (expression ratio = ER) and the mean of the four replicate spots for each cDNA (2 for each gene = ER1, ER2) were calculated. ERs were subjected to a t-test to determine if the values differed significantly from 1. A transcript was defined as being differentially regulated if both of the following criteria were fulfilled: 1) the final ER (ER1+ER2)/2) was equal to or exceeded the arbitrary thresholds \[≤ 0.81 (log~2~0.81 = -0.3) for down-regulated genes or ≥ 1.24 (log~2~1.24 = 0.3) for up-regulated genes\]; 2) both ER1 and ER2 were significantly different from 1 as evaluated by t-tests to control for ER-variance and ER-sample size. An arbitrary threshold was utilized for two reasons: first, to account for normalization errors, and second, to account for the fact that replicate data did not result from repeated hybridizations with the same RNAs but from repeated probe spotting. To evaluate our criteria, we hybridized three arrays with the same cDNA pools and found that 210 of 234 genes (84%) had the same regulation identified by the criteria described above. Of the 30 genes that did not show consistent regulation between the two repeated hybridizations, 24 had the same direction in mean ER but did not meet the statistical requirements for a significant change. These 3 replicate arrays were located together in both the cluster analysis (Fig. [3](#F3){ref-type="fig"}) as well as the PCA (Fig. [4](#F4){ref-type="fig"}). To further estimate the variance of ERs, the mean coefficient of variation (CV) was calculated for each of the genes (gene CV) and each of the arrays (array CV). Gene CVs were obtained by calculating the mean of the individual CVs for each gene on each array; they were used to evaluate the effects of PCR product quality and the thresholds were used to determine significant expression. The quality for each gene was regarded as too low when its mean CV (mean of all 73 arrays) was higher than 0.3. Gene CVs are not influenced by the absolute expression values and reflect the variation across replicate spots on a given array. Array CVs were calculated as the mean of all individual gene CVs for each array and were used to evaluate array and cDNA quantity and quality that was hybridized to the arrays. Since the 73 arrays analyzed in this study were hybridized over 8 months after the arrays were spotted, we used array CV to assess array ageing. A cluster analysis of 35 arrays was performed based on Ward\'s method and the squared euclidian distance \[[@B23],[@B24]\]. To evaluate the appropriate model for the description of the gene distribution, a Detrended Correspondence Analysis (DCA) was performed. The given dimensionless value for the length of gradient of the first ordination axis was \< 1.8, which indicated that the values were better fitted by a linear (lg \< 3) than a unimodal (lg \> 4) distribution model \[[@B25]\]. Therefore, a PCA based on a linear model was chosen to compare gene expression within the microarrays. PCA was performed on log-transformed mean expression ratios of all transcripts from a sample of 35 arrays. Scaling was focused on inter-array distances. Four genes of known function and two of unknown function -- from the PCA analysis (Fig. [4](#F4){ref-type="fig"}), these proved to be good discriminators of the arrays -- were selected to calculate a gene-specific \'norm of reaction\'. For this analysis, mean expression ratios for both PCR-fragments of each of these 6 genes over all arrays were calculated and hierarchically ordered on a log-based scale. For one of these genes (XTH), the error structure on a non-log scale is presented (Fig. [1](#F1){ref-type="fig"} and inset). To test for differences between the groups of different PCR qualities, Kruskal-Wallis ANOVA on Ranks was used. Test statistics and cluster analyses were performed with SPSS 11.0, PCA was carried out with the Canoco 4.5 package \[[@B25]\]. Author\'s contributions ======================= MH carried out the analysis of the microarray data, KG supervised the molecular work. ITB conceived of and coordinated the project, and ITB and MH wrote the manuscript. Acknowledgements ================ We thank T. Hahn and S. Kutschbach for technical assistance in microarray hybridization and data analysis. This research was supported by the Max-Planck-Gesellschaft.	PubMed Central	2024-06-05T03:55:48.384395	2004-9-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522747/", "journal": "BMC Ecol. 2004 Sep 7; 4:13", "authors": [ { "first": "Matthias", "last": "Held" }, { "first": "Klaus", "last": "Gase" }, { "first": "Ian T", "last": "Baldwin" } ] }
PMC522748	Background ========== The safety of genetically engineered foods must be assessed before authorities in most nations will consider granting market approval. An important issue in current food safety assessment is the evaluation of the potential allergenicity of food derived from biotechnology. Since many food allergens are proteins, introduction of a new (\"foreign\") protein in food by genetic engineering can in theory cause allergic reactions. Therefore the allergenicity of novel proteins needs to be assessed. Potential allergenicity of a protein is a complex issue and various tests can be used for prediction, including bioinformatics, in vitrodigestibility and binding of antisera of allergic patients. A step-by-step procedure to assess allergenicity is described by the Codex alimentarius and the FAO/WHO consultation group \[[@B1],[@B2]\]. An important step in this procedure is to use bioinformatics to determine whether the primary structure (amino acid sequence) of a given transgenic protein is sufficiently similar to sequences of known allergenic proteins. The recommended procedure \[[@B1]\] to establish the possibility of allergenicity is to: \(1) Obtain the amino acids sequences of known allergens in protein databases in FASTA format (using the amino acids from the mature proteins only, disregarding the leader sequences, if any). \(2) Prepare the complete set of 80-amino acid length sequences derived from the query protein (again disregarding the leader sequence, if any). \(3) Compare each of the sequences of (2) with all sequences of (1), using the program FASTA \[[@B3]\] with default settings for gap penalty and extension. According to the Codex alimentarius \[[@B2]\], potential allergenicity should be considered, when there is either: \(a) More than 35 % similarity over a window of 80 amino acids of the query protein with a known allergen. \(b) A stretch of identity of 6 to 8 contiguous amino acids. This procedure is described in more detail by the expert consultation and the Codex Alimentarius. Potential allergenicity requires further testing of the protein with panels of patient sera and possibly animal exposure tests \[[@B1],[@B2]\]. Construction and content ======================== Three allergen databases were created, one derived from SwissProt \[[@B4]\] and one from the WHO-IUIS allergen list \[[@B5]\]. A third database is a non-redundant combination of the other two. The databases were created by extracting all proteins from public databases; SwissProt (version 44.1, July 5 2004, \[[@B4]\]), PIR \[[@B6]\] and GenPept <http://www.ncbi.nlm.nih.gov>. Leader sequences were, if annotated, trimmed from the sequence. The SwissProt allergen list contains 334 mature protein sequences, while the WHO-IUIS allergen list (version June 7, 2004) contains 632 sequences (correcting for three internal duplications). These two databases contain 236 duplicate entries. The non-redundant combined database contains 730 sequences (Figure [1](#F1){ref-type="fig"}). Allermatch™ is build around the FASTA package (version 3.4t21; <ftp://ftp.virginia.edu/pub/fasta/>, \[[@B3]\]) running with default parameters (ktup = 2, matrix = Blosum50, Gap open = -10, Gap extend = -2). The Allermatch™ analysis tool and the web interface are written in Python and run on a Suse L Linux Enterprise server with an Apache web server (version 1.3.26). Allermatch™ provides two search methods (mode 1 & 2) corresponding with the FAO/WHO guidelines described above and a third method (mode 3) is provided as an extra tool. The outline of the application is schematically presented in Figure [2](#F2){ref-type="fig"}. Mode 1: Sliding window approach ------------------------------- The query protein sequence is divided into 80 amino acid (aa) windows using a sliding window with steps of a single residue. Each of these windows is compared with all sequences in the allergen database of choice. All database entries showing a similarity higher than a configurable threshold percentage (default is 35%) to any of the 80 aa query sequence windows are flagged. Upon completion of the analysis, a table is shown with all flagged database entries. Per entry, the highest similarity score is given, as well as the number of windows having a similarity above the cut-off percentage. For each allergen database entry identified, more detailed information on the similarity between the allergen and query sequence can be retrieved, such as those areas of both proteins within all 80 aa windows scoring above the cut-off percentage. The similarity score calculated by FASTA can apply to stretches smaller than 80 aa, Allermatch™ converts such a similarity score to an 80 aa window. For example, 40% similarity on a stretch of 40 aa converts to 20% similarity on an 80 aa window. Mode 2: Wordmatch ----------------- This method looks for short sub-sequences (words), which have a perfect identity with a database entry. The wordsize is configurable (default is 6 aa). The output given is similar to the output given by Mode 1. All database entries with at least one hit are listed and for each of these, more detailed information can be retrieved upon request. Mode 3: full FASTA alignment with an Allermatch™ allergen database ------------------------------------------------------------------ The Allermatch™ webtool also offers a full alignment of the query sequence with either of the allergen databases using FASTA. Although this full alignment is currently not required by the FAO/WHO guidelines, the full alignment of protein sequences helps positioning of regions of potential allergenicity in the whole primary structure of the protein. The FASTA output is parsed and information from the allergen database is added and presented. Utility and discussion ====================== To examine the predictive performance of the FAO/WHO criteria for potential allergenicity, we have performed two tests. The first test determines the percentage of false negative and the second test assesses the amount of false positives. Both tests are performed with standard settings; for the sliding window approach an 80 amino acid window with a 35% similarity cutoff is used and for the wordmatch approach 6, 7 and 8 aa word sizes are tested. The false negative error-rate is estimated by a leave-one-out method, testing all sequences in each Allermatch™ database against that database with the tested sequence excluded. Each sequence not resulting in a hit is considered a false negative. The results of each method/database combination are summarized in Table [1](#T1){ref-type="table"}, column 1. The results show that the number of false negatives decreases when a larger database of allergen sequences is used. This may (partly) be explained by an increased proportion of similar, but not equal, sequences in the larger databases, such as isoallergens listed by WHO-IUIS. In examining the results, various sequences were observed that were not able to produce a hit (data not shown) due to their short length, since a perfect hit on a sequence shorter than 28 amino acids cannot convert to a 35% hit on an 80 amino acid window. Column 2 of the same table shows the corrected false negative rate after exclusion of these sequences. Also after this correction the wordmatch with 6 amino acids method shows lower numbers of false negatives than the sliding window approach. It is clear, however, that in case of short protein sequences the sensitivity of the sliding window methods is reduced. In the second test, we assess the odds of a false positive by testing 12 protein sequences known to be non allergenic. This is based on non-reactivity of these proteins towards IgE-sera of allergy patients or on the inability to cause IgE-responses in experimental animals (Table [2](#T2){ref-type="table"}). It should be noted that such data are only available for a limited number of proteins, which accounts for the size of this dataset. Each of these 12 sequences was tested against all databases with all methods. Each non-allergenic sequence resulting in a hit is considered a false positive (Table [1](#T1){ref-type="table"}, column 3). The number of false positives grows with the database size, as is to be expected: the chance of a random hit increases with a larger database. In contrast to the false negative hit rates the sliding window method gives the lower error rate. This test might, however, overestimate the number of false positives. A number of these non-allergens are related to and display similarities with their allergenic counterparts, i.e.T1 (related to Bet v 1), human serum albumin (related to animal serum albumins), and human heat shock protein 70 (similar to heat shock proteins from fungi and other allergens). A selection of unrelated, non-allergenic proteins is therefore likely to give a lower false positive rate. Caution should be taken in interpreting these false hit rates. The used methods might perform differently with other sets of proteins. For example, a member of a completely novel group of valid allergens is likely to generate a false negative result. The imperfect results show here agree with literature where the FAO/WHO methods for sequence comparisons are also shown to lack full predictive capability \[[@B7]-[@B9]\]. Interestingly, the results show that there is a balance between false positives and negatives when increasing the threshold level for short exact matches from 6 to 8 amino acids, with the number of false positives sharply decreasing at 8 amino acids (Table [1](#T1){ref-type="table"}). The outcomes of these tests therefore need to be further refined by checking for the presence of potential IgE-epitopes as recommended by Kleter and Peijnenburg \[[@B7]\], as well as combined with results of other assays as recommended by the Codex. Other methods to decrease false hit rates may also be considered \[[@B8],[@B9]\]. We plan to implement such supplementary methods in the future to support the Codex based predictions of potential allergenicity. The prediction of potential allergenicity by primary sequence comparison depends on the quality of the data used for comparison. Addition of a non-allergenic or poorly annotated protein to any of the Allermatch™ allergen databases would obviously result in undesired false positives and should be prevented. A workable strategy could be to use multiple databases, i.e.a database based on SwissProt\'s list of allergens, which contains well-annotated sequences from SwissProt, simultaneously with a larger database based on the WHO-IUIS list, which contains possibly less well annotated sequences from other protein databases, such as GenPept. For example, a number of protein accessions in the WHO-IUIS database do not mention the presence of signal- and/or pro-peptides, where removal of such peptides is essential to prevent false positives. Users of Allermatch™ should, at all times, take into account the possibility of a false positive or negative, for example by checking original data (accessions, clinical literature) and confirm results, before arriving at conclusions. To prevent false positives as much as possible, one should choose for the well-annotated SwissProt database. To prevent false negatives, the combination of the larger WHO-IUIS database with that of SwissProt is more appropriate. Updates to the SwissProt and WHO-IUIS allergen lists will be incorporated in the Allermatch™ databases on a regular basis. Several other websites in the public domain offer sequence alignment facilities that support the prediction of potential allergenicity, such as SDAP \[[@B10],[@B11]\], AllerPredict \[[@B12]\] and Farrp \[[@B13]\]. These websites offer search algorithms that find contiguous similar amino acids between a query sequence and database sequences (SDAP, AllerPredict) and more than 35% identity in alignments (SDAP, AllerPredict), as well as a general FASTA of a query protein sequence against the database (SDAP, Farrp). Conclusions =========== Allermatch™ is an efficient and comprehensive webtool that combines all bioinformatics approaches required to assess the allergenicity of protein sequences according to the current guidelines in the Codex. The application will be kept up to date with the FAO/WHO criteria and the SwissProt and WHO-IUIS allergen lists. It will be extended with other, supplementary methods to support and refine the prediction of allergenicity. Availability and requirements ============================= Allermatch™ is platform independent and accessible using any Netscape 4+ compatible webbrowser at <http://allermatch.org>. Authors\' contributions ======================= MF developed and implemented the Allermatch™ webtool. HN provided the domain name registration and advised in the web site development. GK and AP provided the scientific background and constructed the sequence databases. JPN and RvH provided time, resources and ample discussion. All authors have read and approved the final manuscript. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A Venn-diagram showing the relationships of the three databases provided by Allermatch™.This figure shows the size and overlap between the SwissProt and WHO-IUIS allergen databases. ::: ![](1471-2105-5-133-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Schematic representation of the Allermatch™ webtool.The user submits a protein sequence of interest to the Allermatch™ webtool and chooses one of the three alignment methods and three databases available. Upon completion the results are formatted and returned to the user. ::: ![](1471-2105-5-133-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Prediction quality of the FAO/WHO methods. ::: 1 2 3 ---------------------- ----------- ------ ----------- ------ ---------- ------ -------- ------ SwissProt Window n.a. 71 / 334 21.3 57 / 320 17.8 3 / 12 25.0 Wordmatch 6 54 / 334 16.2 n.a. n.a. 7 / 12 58.3 7 69 / 334 20.7 n.a. n.a. 6 / 12 50.0 8 78 / 334 23.4 n.a. n.a. 3 / 12 25.0 WHO-IUIS Window n.a. 99 / 632 15.7 78 / 611 12.8 4 / 12 33.3 Wordmatch 6 58 / 632 9.2 n.a. n.a. 9 / 12 75.0 7 98 / 632 15.5 n.a. n.a. 8 / 12 66.7 8 117 / 632 18.5 n.a. n.a. 3 / 12 25.0 SwissProt & WHO-IUIS Window n.a. 101 / 730 13.8 77 / 706 10.9 5 / 12 41.7 Wordmatch 6 55 / 730 7.5 n.a. n.a. 9 / 12 75.0 7 95 / 730 13.0 n.a. n.a. 8 / 12 66.7 8 115 / 730 15.8 n.a. n.a. 3 / 12 25.0 The number and percentage of false negative and false positive results are shown here for all FAO/WHO recommended method/database combinations. Result set 1 describes the number of false negatives observed in a leave-one-out approach. The next result set (2) shows the same results but corrected for those sequences that were not able to generate a hit against itself due to the short length of the sequence. The last result set (3) shows the observed number of false positives when testing 12 non-allergenic sequences with the Allermatch™ webtool. Each of the result sets consists of two columns; the first column shows the number of erroneous hits and the total number of sequences in this set. The second column shows the percentage of erroneous hits. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Sequences used for the negative control ::: Protein Host organism Evidence for non-allergenicity Accession Reference --------------------------------- ---------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------ ----------- Amaranth seed albumin Amaranthus hypochondriacus IgG-response, but no raised IgE-levels, after administration (intranasal and intraperitoneal) of amaranth seed albumin to mice GenPept CAA77664 \[14\] T1 Catharanthus roseus No reaction of recombinant T1 in IgE-sera binding, basophile histamine release, and skin prick testing using patients allergic to the related birch pollen allergen Bet v 1 Not applicable \[15\] Mite ferritin heavy chain Dermatophagoides pteronyssinus Reaction of mite ferritin with IgG, but not with IgE, of sera from patients allergic to house dust mite GenPept AAG02250 \[16\] Maltose binding protein Escherichia coli No reaction with IgE-sera from patients allergic to natural rubber latex (maltose binding protein used as part of fusion proteins with latex allergens) SwissProt P02928 \[17\] Human serum albumin Homo sapiens No reaction of human serum albumin with IgE-sera of patients allergic to cat- and porcine-serum albumin SwissProt P02768 \[18\] Human heat shock protein 70 Homo sapiens No reaction of human heat shock protein 70 with IgE-sera of patients allergic to heat shock protein 70 from Echinococcus granulosus SwissProt P08107 \[19\] Human beta-2-glycoprotein I Homo sapiens Presence of IgM antibodies, but not of IgE antibodies, directed against human beta-2-glycoprotein I in sera from atopic eczema/dermatitis patients SwissProt P02749 \[20\] Guayule rubber particle protein Parthenium argentatum No cross-reactivity between proteins from guayule and latex using IgE-sera from patients allergic to latex Swissprot Q40778 \[21\] Purple acid phosphatase 1 Solanum tuberosum Stimulation of IgG-, but no or only low stimulation of IgE-antibodies following administration of potato acid phosphatase to mice (oral and intraperitoneal) TrEMBL Q6J5M7 \[22\] Purple acid phosphatase 2 Solanum tuberosum See above TrEMBL Q6J5M9 \[22\] Purple acid phosphatase 3 Solanum tuberosum See above TrEMBL Q6J5M8 \[22\] Potato lectin Solanum tuberosum Stimulation of IgG-, but no or only low stimulation of IgE-antibodies following administration of potato lectin to mice (intraperitoneal) TrEMBL Q9S8M0 \[23\] :::	PubMed Central	2024-06-05T03:55:48.387176	2004-9-16	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522748/", "journal": "BMC Bioinformatics. 2004 Sep 16; 5:133", "authors": [ { "first": "Mark WEJ", "last": "Fiers" }, { "first": "Gijs A", "last": "Kleter" }, { "first": "Herman", "last": "Nijland" }, { "first": "Ad ACM", "last": "Peijnenburg" }, { "first": "Jan Peter", "last": "Nap" }, { "first": "Roeland CHJ", "last": "van Ham" } ] }
PMC522749	Background ========== Mortality and morbidity due to cardiovascular diseases are frequent in patients with diabetes mellitus and high prevalence of diabetes and cardiovascular disease, also, are observed in patients with end-stage renal disease treated by renal replacement therapy, either renal transplantation (RT) and dialysis \[[@B1]\]. Although uremia is typically associated with impaired glucose metabolism via multiple mechanisms \[[@B2]-[@B4]\], hemodialysis improves, although not completely, the uremic induced glucose impairment \[[@B5]-[@B7]\]. Impaired glucose metabolism is also a common and an important problem after RT. By improvement of immunosuppression after RT, the incidence of post transplant diabetes (PTDM) has been decreased from 41% to 2.5% \[[@B8],[@B9]\]. Although we routinely screen and treat only full-blown diabetes at the post transplant periods, an overlooked aspect is the impaired glucose tolerance, which may be a risk factor to induce atherosclerosis. Impaired glucose tolerance de novo, may be a risk factor of post-transplantation mortality and morbidity \[[@B10]\]. Although increased levels of glycosylated hemoglobin (HbA1C) and lipid concentrations have been shown in hemodialysis patients \[[@B11]\] and renal transplant recipients \[[@B12]\] with diabetes, their impairment is not clear in the both groups with impaired glucose tolerance without apparent diabetes mellitus. In this study we investigated glucose tolerance and lipid profiles in non-diabetic hemodialysis and renal transplant patients. Methods ======= We selected fifty five RT recipients with more than one year of good renal allograft function (serum Cr \< 1.5 mg/dl), under conventional triple therapy composed of cyclosporine A (CsA), azathiopurine and prednisolone. Their allograft sources were living donors. Fifty five stable HD patients and another fifty five healthy controls (C), were also enrolled in this study. The mean age (39 ± 7 years), sex (F/M ratio was 33/22), body mass index (BMI) 24.7 ± 1.28 kg/m^2^) were similar in the three groups (see table [1](#T1){ref-type="table"}). Patients with diabetes mellitus and propranolol consumers were excluded. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographical, biochemical, hematological and therapeutical factors in hemodialysis patients and renal transplant recipients. ::: HD RT Control --------------------------------------------- -------------- ------------ ------------ Age (years) 48 ± 3 46 ± 4 47 ± 4 Male/female ratio 33/22 33/22 33/22 BMI (Kg/m^2^) 24.6 ± 1.4 23.8 ± 1.2 23.6 ± 1.3 Cholesterol (mg/dl) 199.2 ± 36.6 269.7 ± 54 190.5 ± 34 Triglycerides (mg/dl) 217.2 ± 55 214.3 ± 13 100.2 ± 18 HgA1C (g/dl) 6.42 ± 0.7 6.2 ± 0.9 5.7 ± 0.7 Hb level 10.9 ± 0.8 12.4 ± 1.1 13.3 ± 0.8 Therapy with vitamin D3 0.5 μg/day (number) 25 \-- \-- Impaired glucose tolerance (number) 12\* 4\\ \-- \Significant correlation with BMI, serum triglycerides and HgA1C \\* Significant correlation with serum cholesterol, CysA concentration and HgA1C ::: The levels of serum triglyceride, cholesterol (measured by enzymatic spectrophotometry)\[[@B13]\], CsA (measured by ELISA in whole blood, only in renal transplant recipients) and glycosylated haemoglobin concentration (Hb A1c) (measured by column chromatography) were measured after 10 hours fasting (in the hemodialysis group, in the early morning before hemodialysis). Fasting blood sugar and the standard 2 hours glucose tolerance test (after ingestion of 75 g of glucose) were detected in the three groups by spectrophotometry. Statistical analysis was performed by Kuruskal wallis, U-Mann Whitney, multiple comparison and regression correlation coefficient tests, using SPSS 10.05. Results ======= On the basis of WHO classification \[[@B14]\], four of our (7.5%) renal transplant recipients and twelve (22%) of the hemodialysis patients had impaired glucose tolerance, i.e. the 2 hour of glucose tolerance test was between 140 and 200 mg/dl. It was more obvious at the end of the second hour of GTT. Although BMI was roughly similar in the three groups (Table [1](#T1){ref-type="table"}), a significant linear correlation was observed between BMI and impaired glucose tolerance only in HD patients (r = 0.4, p = 0.05) (fig [1](#F1){ref-type="fig"}), but not in the RT recipients. The glucose tolerance (especially at the first hour) in the HD patients had a significant linear correlation with the level of serum triglycerides (r = 0.87, p = 0.001) (Fig [2](#F2){ref-type="fig"}). Serum triglyceride concentration was 217.2 ± 55 mg/dl in HD vs. 214.3 ± 13 mg/dl in RT and 100.2 ± 18 mg/dl in C groups, (p = 0.001). On the other hand the four RT recipients with IGTT (i.e. 100% of RT recipients with IGTT) had the higher serum cholesterol levels (308.4 ± 24.4 mg/dl)) compared with the remaining RT recipients with normal GTT (248.7 ± 55.6 mg/dl) with p = 0.031 (table [2](#T2){ref-type="table"}). The mean of serum cholesterol was 269.7 ± 54 mg/dl in RT vs. 199.2 ± 36.6 mg/dl in HD and190.5 ± 34 mg/dl in C groups (p = 0.0001). A linear correlation was observed between impaired GTT and both of the serum Cyclosporine level (r = 0.9, p = 0.001) and HbA1c in RT recipients (Fig [3](#F3){ref-type="fig"}). The mean of HbA1c was 6.2 ± 0.6 gr/dl in the RT recipients with normal GTT vs. 4.34 ± 0.26 g/dl in the RT recipients with IGGT (p \< 0.001, table [2](#T2){ref-type="table"}). The later correlation was also observed in HD patients, in whom the mean of HbA1C level was 6.4 ± 0.7 gr/dl in the group (r = 67, p = 0.001). In contrast of a close relationship of IGTT and higher HbA1c, the gender, age, times after transplantation and BMI did not impact on IGTT in RT recipients. Although in logistic regression analysis higher serum level of cyclosporine was correlated with increased GTT impairment, we could not evaluate the implication of corticosteroids on this test, because all of the 55 RT recipients were received prednisolone at a doses of 5 to 10 mg/day. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### GTT has a linear relationship with BMI in hemodialysis patients. Impairment of GTT is more significant in the hemodialysis patients with higher BMI. Gtt2 = glucose tolerance test at the second hours of 75 gr oral glucose. bmih = body mass index in hemodialysis patients ::: ![](1471-2369-5-11-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### The glucose tolerance in the HD patients had a significant linear correlation with the level of serum triglycerides. gttd1= glucose tolerance test in dialysis patients, tgh= serum concentration of triglyceride in hemodialysis patients. ::: ![](1471-2369-5-11-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Cyclosporine level and HbA1c have correlations with the IGTT in RT recipients ▲ = Serum Cyclosporine level ○ = HbA1c concentration ::: ![](1471-2369-5-11-3) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Impaired glucose tests in HD and RT recipients have higher values of serum triglyceride, serum cholesterol and cyclosporine concentration than patients with normal glucose tolerance tests. ::: no. of cases Serum Triglyceride(mg/dl) Serum Cholesterol(mg/dl) HbA1c (gr/l) Cyclosporine (mg/dl) ------------------------------- -------------- --------------------------- -------------------------- ----------------- ---------------------- RT recipients with IGTT 4 231.4 ± 150 308.4 ± 24.4 7.34 ± 0.26 320.4 ± 36.6 RT Recipients With normal GTT 51 201 ± 75 248.7 ± 55.6 6.2 ± 0.6 295.1 ± 29 *P = 0.59* *P = 0.02* *P = 0.001* *P = 0.2* HD patients with IGTT 12 272.1 ± 41.3 201.1 ± 39 7 ± 1 HD patients with normal GTT 43 195.9 ± 45 198.5 ± 36.8 5.9 ± 0.7 *P = 0.001* *P = 0.87* *P = 0.007* Controls 55 100.2 ± 18 190.5 ± 34 5.7 ± 0.7 ::: Discussion ========== Impaired glucose tolerance occurs in about 50% of patients with chronic renal failure (CRF) patients. It is due to multiple factors, which the two most important of them being insulin resistance at target organs and impaired insulin secretion from the pancreas \[[@B15]\]. Insulin sensitivity would be reduced by up to 60% in non-diabetic patients with CRF before dialysis \[[@B16]\]. Marked improvement in insulin sensitivity and consequently glucose tolerance has been reported in non-diabetic patients after 10 weeks of HD, although they are not completely returned to normal \[[@B15]\]. Thereby, impaired glucose tolerance during HD is secondary to non-effective removable toxins by HD compared with peritoneal dialysis. In the latter more effective removal of middle molecule toxins causes better glucose tolerance, although glucose rich dialyzet solution is used \[[@B16]\]. The other causes of impaired glucose tolerance in HD patients may be secondary to metabolic disturbances, such as anemia \[[@B17]\], malnutrition \[[@B18]\] and vitamin D3 deficiency \[[@B19]\]. Although all of our HD patients had normochromic-normocytic anemia, the severity was not proportionate with impaired glucose tolerance (The data has not been shown). The patients were well nourished and were under treatment with daily oral vitamin D3 (Rocaltrol), 0.5 micrograms per day. So malnutrition and vitamin D3 deficiency could not to contribute to impaired glucose tolerance in our HD patients. Impaired glucose tolerance was also observed in 7.5% of our RT recipients. All of the presumed risk factors for post transplant diabetes mellitus such as old age \[[@B18]\], family history of any known diabetes mellitus in their first relatives\[[@B21]\], cadaveric allografts \[[@B22]\] and obesity did not exist in the patients. Previously Boudreaux et al. \[[@B23]\] reported that those patients who weighed more than 70 kg had a higher incidence of post transplant diabetes mellitus (PTDM). A relative risk of 1.4 for developing PTDM for every 10 kg increase in body weight more than 60 kg has been shown \[[@B12]\]. Although in our study obese patients (BMI \> 30 kg/m2) were not included in the both groups, a correlation was observed between impaired glucose tolerance and higher BMI in our HD patients. In RT recipients, the major risk factor for impaired glucose tolerance was immunosuppressive therapy. Through using higher doses of CsA and corticosteroids, PTDM was previously more common, but the complication has been decreased to 2--5% in FK506-based immunosuppressive protocols \[[@B24],[@B25]\]. Although this relatively uncommon complication is a major cause of post-transplant mortality and morbidity, even minor glucose intolerance is associated with an increased long-term risk for cardiovascular disease \[[@B26]\]. The importance of impaired glucose tolerance should not be underestimated in these patients with high risk of atherosclerosis. Hyperlipidemia, another risk factor for atherosclerosis, on one hand accompanies the impaired glucose tolerance observed in the HD and RT patients and on the other hand increases the risk of atherosclerosis induced by impaired glucose tolerance. As reported previously, a tendency to higher pre-transplantation serum triglyceride concentration was associated with post-transplantation impaired glucose tolerance \[[@B27]\]. Hypertriglyceridemia is common complication in dialysis patients. In non-transplant populations it is regarded (along with low HDL cholesterol levels) as a prominent feature of insulin resistance syndrome, and also is a cardiovascular risk factor in organ transplant recipients \[[@B28]\]. Our study confirmed the relationship between impaired glucose tolerance and triglyceride levels in HD patients, and between impaired glucose tolerance and cholesterol levels in RT recipients. The latter was also accompanied by a higher level of HgA1C. Commonly used tests of HgA1C may be unreliable in patients with end-stage renal disease because of the presence of anemia, shortened red blood cell survival, and assay interferences from uremia. But HgA1C in the range of 6% to 7%, as was found in our study, estimates glycemic control within the range of patients without severe renal impairment \[[@B1]\]. So in the range of mild to moderate increased HgA1C in HD and uremic patients, it would be a reliable marker of impaired glucose tolerance. Conclusions =========== There was increased HgA1C and impaired glucose tolerance in HD and RT patients. This was accompanied by hyperlipidemia in HD patients (with hypertriglyceridemia) and RT recipients (with hypercholesterolemia). The impact upon the progression of atherosclerosis needs more study in haemodialysis and renal transplant populations at a long term follow up. Competing interests =================== None declared. Authors\' contributions ======================= *HAreviewed the literatures and wrote the manuscript and also helped to do statistical analysis, ANperformed GTT and the other biochemical markers, MNparticipated as coordinator between laboratory and clinic, HTK*selected the patients and collected data Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2369/5/11/prepub> Acknowledgments =============== We acknowledge from laboratory division of Emam hospital in Tabriz university of medical sciences for its cooperation	PubMed Central	2024-06-05T03:55:48.389416	2004-9-8	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522749/", "journal": "BMC Nephrol. 2004 Sep 8; 5:11", "authors": [ { "first": "Hassan", "last": "Argani" }, { "first": "Alireza", "last": "Noorazarian" }, { "first": "Mohammad", "last": "Rahbaninobar" }, { "first": "Mohammad", "last": "Noori" }, { "first": "Hamid T", "last": "Khosroshahi" } ] }
PMC522750	Background ========== One of the major clinical problems is how to best evaluate and manage the increasing numbers of patients infected with the hepatitis C virus (HCV) \[[@B1]\]. Liver biopsy is still recommended in most patients \[[@B2],[@B3]\]. However, numerous studies strongly suggest that due to the limitations \[[@B4]-[@B6]\] and risks of biopsy \[[@B7]\], as well as the improvement of the diagnostic accuracy of biochemical markers \[[@B8],[@B9]\], liver biopsy should no longer be considered mandatory. Among the non-invasive alternatives to liver biopsy \[[@B10]\], several studies have demonstrated the predictive value of two combinations of simple serum biochemical markers in patients infected with HCV: FibroTest (FT; Biopredictive, Paris, France; HCV-Fibrosure, Labcorp, Burlington, USA) for the assessment of fibrosis; and ActiTest (AT; Biopredictive, Paris, France) for the assessment of necroinflammatory activity (necrosis) \[[@B8],[@B9],[@B11]-[@B21]\]. Similar results have not been obtained with other diagnostic tests \[[@B10]-[@B17]\]. Since September 2002 these tests (FT-AT) have been used in several countries as an alternative to liver biopsy. In a recent systematic review, it was concluded that these panels of tests might have the greatest value in predicting fibrosis or cirrhosis \[[@B10]\]. It was also stated that biochemical and serologic tests were best at predicting no or minimal fibrosis and at predicting advanced fibrosis/cirrhosis, and were poor at predicting intermediate levels of fibrosis \[[@B10]\]. The aim of this study was to summarize the diagnostic value of these tests by an overview of the scientific literature and to respond to the following frequently asked questions by performing original new analyses: 1) what is the range of the FT-AT diagnostic values across the different studies? 2) What are the base evidence comparisons between FT-AT and other published biochemical markers? 3) Are there differences in diagnostic values according to HCV genotype or viral load? 4) Are there differences between the FT-AT diagnostic values according to stages and grades? -- In other words, is FT better at predicting no or minimal fibrosis (F0 vsF1) or advanced fibrosis/cirrhosis (F3 vsF4) than at predicting intermediate levels of fibrosis (F1 vsF2)? And 5) what is the conversion between FT-AT results and the corresponding fibrosis stages and necrosis grades? Results ======= Analysis of the literature -------------------------- Between February 2001 and March 2004, a total of 16 publications \[[@B8],[@B9],[@B11]-[@B21],[@B24]-[@B26]\] and 4 abstracts \[[@B27]-[@B30]\] without corresponding publications were identified. Diagnostic value of FT-AT among published studies ------------------------------------------------- For 12 groups of patients detailed in 6 publications \[[@B8],[@B11],[@B12],[@B14],[@B19],[@B26]\], it was possible to assess the prevalence of significant fibrosis and the FT area under receiver operating characteristics curve (AUROC) values, as well as the sensitivity and specificity for the 4 different FT cut offs (Table [1](#T1){ref-type="table"}). For the diagnosis of significant fibrosis by the METAVIR scoring system, the AUROC ranged from 0.73 to 0.87, significantly different from random diagnosis in each study (Table [1](#T1){ref-type="table"}), in meta-analysis (mean difference in AUROC = 0.39, random effect model Chi-square = 529, P \< 0.001) (Figure [1](#F1){ref-type="fig"}, upper panel), or after pooling data in the integrated database (Table [2](#T2){ref-type="table"}). For the cut off of 0.31, the FibroTest negative predictive value for excluding significant fibrosis (prevalence 0.31) was 91% (Table [2](#T2){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Summary of the diagnostic value of FibroTest for the staging of hepatic fibrosis and comparisons with hyaluronic acid, the Forns Index and the APRI Index in patients with chronic hepatitis C, from the published studies. ::: ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- First author N\* Methodology Marker Stage/Prevalence AUROC SE Cut off Sensitivity Specificity --------------------- --------- -------------------------- ------------------- ---------------------- -------------- ------------- ----------------- ----------------- Imbert-Bismut, 2001 189 Prospective\ FibroTest F2F3F4 / 0.38 0.84 (0.03) 0.10\ 0.97\ 0.24\ Single center\ 0.30\ 0.79\ 0.65\ First year cohort 0.60\ 0.51\ 0.94\ 0.80 0.29 0.95 Imbert-Bismut, 2001 134 Prospective\ FibroTest F2F3F4 / 0.45 0.87 (0.03) 0.10\ 1.00\ 0.22\ Single center\ 0.30\ 0.87\ 0.59\ Validation cohort 0.60\ 0.70\ 0.95\ 0.80 0.38 0.97 Poynard, 2001 165 Retrospective\ FibroTest F3F4 Knodell / 0.32 0.74 (0.03) 0.10\ 0.96\ 0.24\ Randomized trial\ 0.30\ 0.81\ 0.65\ Multicenter 0.60\ 0.50\ 0.92\ 0.80 0.13 0.98 Poynard, 2001 165 Retrospective\ Hyaluronic F3F4 Knodell / 0.32 0.65 (0.03) 20\ 0.81\ 0.39\ Randomized trial\ 40\ 0.47\ 0.65\ Multicenter 100 0.23 0.91 Poynard, 2003 352 Retrospective\ FibroTest F2F3F4 / 0.39 0.73 (0.03) 0.10\ 0.97\ 0.08\ Randomized trial\ 0.30\ 0.86\ 0.45\ Multicenter\ 0.60\ 0.50\ 0.79\ Before treatment 0.80 0.20 0.95 Poynard, 2003 352 Retrospective\ FibroTest F2F3F4 / 0.32 0.77 (0.03) 0.10\ 0.98\ 0.15\ Randomized trial\ 0.30\ 0.85\ 0.39\ Multicenter\ 0.60\ 0.46\ 0.81\ After treatment 0.80 0.16 0.97 Rossi, 2003 125 Prospective\ FibroTest F2F3F4 / 0.38 0.74 (0.05) 0.10\ 0.92\ 0.29\ Multicenter\ 0.30\ 0.75\ 0.61\ Non-validated analyzers 0.60\ 0.42\ 0.94\ 0.80 0.22 0.96 Myers, 2003 130 Retrospective\ FibroTest F2F3F4 / 0.45 0.86 (0.04) 0.10\ 0.98\ 0.17\ Single center\ 0.30\ 0.90\ 0.60\ HCV-HIV Co-infection 0.60\ 0.66\ 0.92\ 0.80 0.34 0.96 Thabut, 2003 249 Retrospective\ FibroTest F2F3F4 / 0.38 0.84 (0.02) 0.10\ 0.98\ 0.22\ Single center\ 0.30\ 0.84\ 0.65\ From Imbert-Bismut, 2001 0.60\ 0.58\ 0.93\ 0.80 0.29 0.95 Thabut, 2003 249 Retrospective\ Forns Index F2F3F4 / 0.38 0.78 (0.03) 1\ 1.00\ 0.04\ Single center\ 3\ 1.00\ 0.26\ From Imbert-Bismut, 2001 6\ 0.55\ 0.86\ 8 0.19 0.97 Le Calvez, 2004 323 Retrospective\ FibroTest F2F3F4 / 0.41 0.83 (0.02) 0.10\ 0.97\ 0.30\ Single center\ 0.30\ 0.81\ 0.66\ From Imbert-Bismut, 2001 0.60\ 0.58\ 0.93\ 0.80 0.33 0.95 Le Calvez, 2004 323 Retrospective\ APRI Index F2F3F4 / 0.41 0.74 (0.03) 0.50\ 0.81\ 0.56\ Single center\ 1.00\ 0.54\ 0.84\ From Imbert-Bismut, 2001 1.50\ 0.36\ 0.91\ 2.00 0.24 0.95 Callewaert, 2004 82 Prospective FibroTest F4 / 0.29 0.89 (0.04) 0.10\ 1.00\ 0.33\ 0.30\ 0.92\ 0.62\ 0.60\ 0.79\ 0.81\ 0.80 0.67 0.92 Callewaert, 2004 82 Prospective Glyco Cirrho Test F4 \\ / 0.29 0.87 (0.04) -0.2\ 1.00\ 0.12\ 0.1\ 0.79\ 0.88\ 0.4\ 0.21\ 0.95\ 0.6 0.17 1.00 ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- \* Number of patients. \\ Compensated ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Meta-analysis of the AUROC observed in published studies of FibroTest diagnostic value.AUROCs were all significantly higher for FibroTest than the random 0.50 value (upper panel) (P \< 0.001). AUROCs of FibroTest were significantly higher then AUROCs of other fibrosis markers (lower panel) (P \< 0.05). ::: ![](1476-5926-3-8-1) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Integrated database, with predictive values for significant hepatic fibrosis according to METAVIR conversion cut offs. Derived from published studies. ::: Integrated database Patient number Marker Stage/Prevalence AUROC (SE) Cut off used for METAVIR stages conversion Sensitivity Specificity Negative predictive value Positive predictive value ------------------------- -------------------- ------------ ---------------------- ---------------- ------------------------------------------------ ----------------- ----------------- ------------------------------- ------------------------------- With Blood Donors 1,570 FibroTest F2F3F4/0.31 0.83 (0.01) 0.21 0.92 0.55 0.94 0.48 0.27 0.87 0.62 0.92 0.51 0.31 0.84 0.68 0.91 0.54 0.48 0.68 0.81 0.85 0.61 0.58 0.56 0.87 0.82 0.67 0.72 0.38 0.95 0.77 0.76 0.74 0.35 0.95 0.76 0.76 0.75 0.33 0.96 0.76 0.78 Without blood donors 1,270 FibroTest F2F3F4/0.38 0.78 (0.01) 0.21 0.92 0.41 0.89 0.49 0.27 0.87 0.48 0.86 0.51 0.31 0.84 0.55 0.85 0.54 0.48 0.68 0.73 0.79 0.61 0.58 0.56 0.83 0.75 0.67 0.72 0.38 0.95 0.70 0.76 0.74 0.35 0.93 0.70 0.76 0.75 0.33 0.94 0.69 0.78 ::: For four groups of patients detailed in two publications \[[@B8],[@B11]\], it was possible to assess the prevalence of significant necrosis and the AT AUROC values, as well as the sensitivity and specificity for 4 different AT cut offs (Table [3](#T3){ref-type="table"}). For the diagnosis of significant necrosis by the METAVIR scoring system, the AUROC ranged from 0.75 to 0.86, significantly different from random diagnosis in each study (Table [3](#T3){ref-type="table"}), in meta-analysis (mean difference in AUROC = 0.29, random effect model Chi-square = 556, P \< 0.001), or after pooling data in the integrated database (Table [4](#T4){ref-type="table"}). For the cut off of 0.36, the ActiTest negative predictive value for excluding significant necrosis (prevalence 0.41) was 85% (Table [2](#T2){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Summary of the diagnostic value of ActiTest for the diagnosis of necroinflammatory hepatic activity (AUROC) in patients with chronic hepatitis C, from the published studies. ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ First author, Year Patient number Methodology Marker Grade/Prevalence AUROC (SE) Cut off Sensitivity Specificity ------------------------ -------------------- ------------------- ------------ ---------------------- ---------------- ------------- ----------------- ----------------- Imbert-Bismut, 2001 189 Prospective\ ActiTest A2A3 / 0.33 0.79 (0.03) 0.10\ 0.99\ 0.07\ Single center 0.30\ 0.91\ 0.42\ 0.60\ 0.70\ 0.75\ 0.80 0.49 0.88 Imbert-Bismut, 2001 134 Prospective\ ActiTest A2A3 / 0.28 0.75 (0.03) 0.10\ 1.00\ 0.07\ Single center\ 0.30\ 0.94\ 0.33\ Validation cohort 0.60\ 0.67\ 0.65\ 0.80 0.42 0.87 Poynard, 2003 352 Retrospective\ ActiTest A2A3 / 0.83 0.75 (0.03) 0.10\ 1.00\ 0.00\ Randomized trial\ 0.30\ 0.90\ 0.38\ Multicenter\ 0.60\ 0.49\ 0.87\ Before treatment 0.80 0.20 0.99 Poynard, 2003 352 Retrospective\ ActiTest A2A3 / 0.39 0.86 (0.02) 0.10\ 0.91\ 0.59\ Randomized trial\ 0.30\ 0.75\ 0.83\ Multicenter\ 0.60\ 0.38\ 0.98\ After treatment 0.80 0.14 0.996 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Integrated database, with predictive values for the diagnosis of significant necroinflammatory hepatic activity according to METAVIR conversion cut offs. Derived from published studies. ::: Integrated database Patient number Marker Grade/Prevalence AUROC (SE) Cut off used for METAVIR stages conversion Sensitivity Specificity Negative predictive value Positive predictive value ------------------------- -------------------- ------------ ---------------------- ---------------- ------------------------------------------------ ----------------- ----------------- ------------------------------- ------------------------------- With Blood Donors 1,570 ActiTest A2A3/0.41 0.85 (0.01) 0.17 0.95 0.55 0.94 0.60 0.29 0.87 0.69 0.88 0.66 0.36 0.81 0.74 0.85 0.69 0.52 0.62 0.86 0.76 0.75 0.60 0.51 0.90 0.72 0.77 0.61 0.50 0.90 0.72 0.78 0.62 0.49 0.91 0.72 0.78 Without blood donors 1,270 ActiTest A2A3/0.51 0.78 (0.01) 0.17 0.95 0.40 0.89 0.62 0.29 0.87 0.55 0.80 0.67 0.36 0.81 0.63 0.76 0.69 0.52 0.62 0.79 0.67 0.75 0.60 0.51 0.85 0.63 0.77 0.61 0.50 0.85 0.62 0.78 0.62 0.49 0.86 0.62 0.78 ::: Comparison of FT-AT diagnostic values with other biochemical markers -------------------------------------------------------------------- In four studies there was a direct comparison in the same patients of FT versusother biochemical markers, including hyaluronic acid \[[@B12]\], the Forns index \[[@B16]\], the APRI index \[[@B17]\] and the GlycoCirrhoTest \[[@B26]\]. All the comparisons were in favor of FT (Table [1](#T1){ref-type="table"}) (Figure [1](#F1){ref-type="fig"}, lower panel), except for the GlycoCirrhoTest, which has a similar AUROC (0.87 vs0.89 for FT) \[[@B26]\]. Integrated database ------------------- A total of 1,570 subjects were included in the integrated database. Of these, 1,270 were patients with chronic hepatitis C who tested PCR positive before treatment and who had had a liver biopsy and METAVIR staging and grading performed. Of these patients, 453 were from our center \[[@B11],[@B14]\], including 130 patients coinfected with HCV and HIV \[[@B14]\]. Eight hundred and seventy (870) patients were from a multicentre study with a total of 398 patients assessed at inclusion and 419 at the end of follow-up six months after treatment; 352 being investigated twice. Three hundred (300) healthy blood donors were also included \[[@B20]\]. Diagnostic value of FT-AT according to HCV genotype and viral load ------------------------------------------------------------------ There was no difference between the AUROC of FT-AT for the diagnosis of significant fibrosis (F2F3F4) (Figure [2A](#F2){ref-type="fig"}) and significant necrosis (A2A3) (Figure [2B](#F2){ref-type="fig"}) between 4 classes of genotype (1, 2, 3 and the rarer genotypes 4, 5, 6 grouped together). There was also no difference between the AUROC of FT-AT of patients with high or low viral loads for the diagnosis of significant fibrosis (Figure [2C](#F2){ref-type="fig"}) or significant necrosis (Figure [2D](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Diagnostic values of FibroTest according to genotype and viral load.[Graph A]{.underline}: AUROCs of FibroTest for the diagnosis of significant fibrosis, according to HCV genotypes. There were no significant differences: Genotype 1, n = 684, AUROC = 0.76, 95% Confidence Interval (95CI) = 0.72--0.79; genotype 2, n = 140, AUROC = 0.79, 95CI = 0.70--0.85; genotype 3, n = 143 AUROC = 0.76, 95CI = 0.67--0.83; other genotype, n = 46, AUROC = 0.72, 95CI = 0.52--0.85. [Graph B]{.underline}: AUROCs of ActiTest for the diagnosis of significant necrosis, according to HCV genotypes. There were no significant differences: Genotype 1, n = 684, AUROC = 0.81, 95% Confidence Interval (95CI) = 0.77--0.84; genotype 2, n = 140, AUROC = 0.90, 95CI = 0.83--0.94; genotype 3, n = 143, AUROC = 0.79, 95CI = 0.71--0.85; other genotype, n = 46, AUROC = 0.76, 95CI = 0.57--0.87. [Graph C]{.underline}: AUROCs of FibroTest for the diagnosis of significant fibrosis, according to serum HCV viral load. There were no significant differences: High viral load, n = 215, AUROC = 0.71, 95% Confidence Interval (95CI) = 0.64--0.78; Low viral load, n = 183, AUROC = 0.73, 95CI = 0.65--0.80. [Graph D]{.underline}: AUROCs of ActiTest for the diagnosis of significant necrosis, according to serum HCV viral load. There were no significant differences: High viral load, n = 215, AUROC = 0.74, 95% Confidence Interval (95CI) = 0.64--0.82; Low viral load, n = 183, AUROC = 0.75, 95CI = 0.65--0.82. ::: ![](1476-5926-3-8-2) ::: Diagnostic value of FT according to the independency of authors --------------------------------------------------------------- Among the 13 published studies of FT (detailed in Table [1](#T1){ref-type="table"}), 9 studies estimated FT and 4 studies compared FT to other non-invasive tests. Among the 9 studies estimating FT, 5 were performed by the same single center (non-independent center), two were performed in totally independent centers, and two were performed in multiple centers, including the non-independent center. The AUROCs for the diagnosis of F2F3F4 versusrandom AUROCs at 0.50, were all significant and similar between these 3 groups in a meta-analysis: mean difference in AUROC = 0.29 (random effect model Chi-square = 549, P \< 0.001), including 0.24 for independent, 0.25 for mixed and 0.36 for dependent studies. In the Callewaert et al. \[[@B26]\] study the AUROC of FT for the diagnosis of F4 was 0.89. Diagnostic value of FT-AT according to stage and grade ------------------------------------------------------ The AUROCs between different stage combinations are given in Table [5](#T5){ref-type="table"}. Between two contiguous stages (one stage difference), the AUROCs were not significantly different and ranged from 0.63 to 0.71. Between patients with a two-stage difference, the AUROCs were not significantly different and ranged from 0.75 to 0.86. Between patients with a three-stage difference, the AUROCs were not significantly different and ranged from 0.87 to 0.95. Between patients with a four- or five-stage difference (blood donors versusF3 or F4, and F0 versusF4), the AUROCs were not significantly different and ranged from 0.95 to 0.99. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Summary of the diagnostic value of FibroTest for the diagnosis of all stage combinations of hepatic fibrosis, according to the AUROCs. ::: F0 F1 F2 F3 F4 BD F0 F0F1 F1F2 F2F3F4 F3F4 -------------------------- -------------------- ------------ ------------ -------------------- -------------------- ----------- ---------- ---------- ------------ ---------- Blood Donor (BD) n = 300 0.71 0.86 *0.95* [0.99]{.underline} [0.99]{.underline} \- 0.84 0.88 0.97 0.99 F0 n = 95 \- F1 n = 688 0.66 \- F2 n = 253 0.82 0.69 \- F3 n = 111 *0.92* 0.80 0.63 \- F4 n = 123 [0.95]{.underline} *0.87* 0.75 0.65 \- BD F0 0.71 0.81 0.92 0.98 0.98 \- F0F1 \- \- 0.71 0.82 0.88 \- \- \- F1F2 0.71 \- 0.69 0.81 0.82 0.84 \- \- \- F2F3 0.85 0.76 \- \- 0.72 0.92 0.80 \- \- F3F4 0.94 0.81 0.81 \- \- 0.98 0.89 0.80 \- F2F3F4 0.83 0.78 \- \- \- 0.95 0.78 \- \- \- BD F0F1 \- \- 0.77 0.87 0.91 \- \- \- 0.83 0.89 The AUROC between all different stage combinations are given. Between two contiguous stages (one- stage difference), the AUROCs are given in bold. Between patients with a two-stages difference, the AUROCs are given in italics. Between patients with a three-stages difference, the AUROCs are given in bold and italics. Between patients with a four- or five-stages difference (blood donors versus F3 or F4, and F0 versus F4), the AUROCs are underlined. Significant differences were observed between AUROCs when there was a two-stage or more difference. ::: The AUROCs between different grade combinations are given in Table [6](#T6){ref-type="table"}. Between two contiguous grades (one grade difference), the AUROCs were not significantly different and ranged from 0.60 to 0.70. Between patients with a two-grade difference, the AUROCs were not significantly different and ranged from 0.75 to 0.86. Between patients with a three-grade difference, the AUROCs were not significantly different and ranged from 0.87 to 0.95. Between patients with a four-grade difference (blood donors versusF3 and F0 versusF4), the AUROCs were not significantly different and ranged from 0.95 to 0.99. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Summary of the diagnostic value of ActiTest for the differential diagnosis of all grades of necroinflammatory hepatic activity, according to the AUROCs. ::: A0 A1 A2 A3 BD A0 A0A1 A1A2 A2A3 ------------------------ ------------ ---------- ---------- ------------------------ ----------- ---------- ---------- ---------- Blood Donor BD n = 300 0.67 0.84 0.96 [0.99]{.underline} \- 0.79 0.89 0.97 A0 n = 185 \- A1 n = 443 0.69 \- A2 n = 370 0.87 0.70 \- A3 n = 272 *0.93* 0.79 0.60 \- A0A1 \- \- 0.70 0.83 \- A1A2 0.77 \- \- 0.70 0.85 \- A2A3 0.89 0.74 \- \- 0.94 0.78 \- A0A1A2 \- \- \- 0.75 \- \- \- \- BD A0A1 \- \- 0.82 0.88 \- \- \- 0.84 The AUROCs between all different grade combinations are given. Between two contiguous grades (one-grade difference), the AUROCs are given in bold. Between patients with a two-grades difference, the AUROCs are given in italics. Between patients with a three-grades difference, the AUROCs are given in bold and italics. Between patients with a four- or five-grades difference (blood donors versus F3 or F4, and F0 versus F4), the AUROCs are underlined. Significant differences were observed between AUROCs when there was a two-grade or more difference. ::: Conversion between FT-AT results and the corresponding fibrosis stage and grade ------------------------------------------------------------------------------- FT-AT is a continuous linear biochemical assessment of fibrosis stage and necroinflammatory activity grade. It provides a numerical quantitative estimate of liver fibrosis ranging from 0.00 to 1.00, corresponding to the well-established METAVIR scoring system of stages F0 to F4 and of grades A0 to A3. Among the 300 controls, the median FT value (± SE) was 0.08 ± 0.004 (95^th^percentile, 0.23) and the median AT value was 0.07 ± 0.004 (95^th^percentile, 0.26). Among the 1,270 HCV-infected patients, the FT conversion was 0.000 -- 0.2100 for F0; 0.2101 -- 0.2700 for F0--F1; 0.2701 -- 0.3100 for F1; 0.3101 -- 0.4800 for F1--F2; 0.4801 -- 0.5800 for F2; 0.5801 -- 0.7200 for F3; 0.7201 -- 0.7400 for F3--F4; and 0.7401 -- 1.00 for F4. (Figure [3A](#F3){ref-type="fig"}). The AT conversion was 0.00 -- 0.1700 for A0; 0.1701 -- 0.2900 for A0--A1; 0.2901 -- 0.3600 for A1; 0.3601 -- 0.5200 for A1--A2; 0.5201 -- 0.6000 for A2; 0.6001 -- 0.6200 for A2--A3; and 0.6201 -- 1.00 for A3 (Figure [3B](#F3){ref-type="fig"}). The conversions are summarized in Figure [4](#F4){ref-type="fig"}. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Conversion between FibroTest and fibrosis stages, and between ActiTest and necroinflammatory activity grades -- Graphs.[Graph A]{.underline}: FibroTest values according to status, from blood donors to patients with cirrhosis (n = 1570). [Graph B]{.underline}: ActiTest values according to status, from blood donors to patients with severe necrosis (n = 1570). F0 = no fibrosis, F1 = portal fibrosis, F2 = some septa, F3 = many septa, F4 = cirrhosis, A0 = no necroinflammatory activity, A1 = minimal activity, A2 = moderate activity, A3 = severe activity. (Consensus conferences recommend treatment in patients with either F2 stage or A2 grade.) Notched box plots showing the relationship between FibroTest and the stage of fibrosis (A) and between ActiTest and the grade of activity (B). The horizontal line inside each box represents the median, and the width of each box the median ± 1.57 interquartile range/√n (to assess the 95% level of significance between group medians). Failure of the shaded boxes to overlap signifies statistical significance (P \< 0.05). The horizontal lines above and below each box encompass the interquartile range (from 25^th^to 75^th^percentile), and the vertical lines from the ends of the box encompass the adjacent values (upper: 75^th^percentile plus 1.5 times interquartile range, lower 25^th^percentile minus 1.5 times interquartile range). ::: ![](1476-5926-3-8-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Conversion between FibroTest and fibrosis stages, and between ActiTest and necroinflammatory activity grades -- Panels.Conversion between FibroTest and fibrosis stages using METAVIR, Knodell and Ishak fibrosis scoring systems (upper panel). Conversion between ActiTest and activity grades using METAVIR, Knodell and Ishak necroinflammatory activity scoring systems (lower panel). ::: ![](1476-5926-3-8-4) ::: Discussion ========== Based on the limitations of liver biopsy and the present overview of the diagnostic value of FT-AT, it seems that these non-invasive markers should be used as a first line assessment of liver injury in patients with chronic hepatitis C. Liver biopsy has three major limitations, which are the risk of adverse events \[[@B2],[@B3],[@B7]\], sampling error \[[@B4]-[@B6]\], and inter- and intra- pathologist variability \[[@B23]\]. An overview of published studies summarizes the risks of liver biopsy as pain (around 30%), severe adverse events (3/1,000) and death (3/10,000) \[[@B2],[@B3],[@B7]\]. Sampling variation is the major cause of variability \[[@B4]-[@B6]\]. In a study of patients with chronic hepatitis C that included only good quality biopsies, 30 of 124 patients (24.2%) had a difference of at least one grade, and 41 of 124 patients (33.1%) had a difference of at least one stage between the right and left lobes \[[@B4]\]. In 18 patients (14.5%), an interpretation of cirrhosis was made in one lobe, whereas stage 3 fibrosis was made in the other \[[@B4]\]. Recently, Bedossa et al. \[[@B6]\] observed very high coefficients of variation (55%) and high discordance rates (35%) for fibrosis staging in biopsies measuring 15 mm in length. The variability significantly improved in biopsies measuring 25 mm in length but was still very high with a 45% coefficient of variation and 25% discordance rate; the minimal variability was reached for biopsies, which were 40 mm in length \[[@B6]\]. Liver biopsy has also potential advantages. Biopsy could be of diagnostic value for other unrecognized liver disease. These events are probably rare in practice, as we observed no such a case in a prospective study of 537 consecutive patients with chronic hepatitis C \[[@B9]\]. For FT-AT it must be realized that the same predictive values were observed for patients coinfected with HIV \[[@B14]\], and in patients with other causes of liver fibrosis such as chronic hepatitis B \[[@B31]\], alcoholic liver disease \[[@B27]\] or non-alcoholic steato-hepatitis \[[@B27]\]. It is possible that biochemical markers such as those described here may provide a more accurate (quantitative and reproducible) picture of fibrogenic and necrotic events occurring within the liver than hepatic biopsy. The greater accuracies of FT-AT, when assessed with biopsy specimens greater than 15 mm versussmaller biopsies, suggest that some discordance between FT-AT and histology were due to biopsy specimen sampling error \[[@B8]\]. Several case reports have observed false negatives of liver biopsy versusbiochemical markers \[[@B8],[@B9],[@B11]\]. The error was attributable to biopsy because there were overt clinical signs of cirrhosis such as esophageal varices, low platelet counts or a dysmorphic liver on ultrasound. In a recent prospective study we estimated that 18% of discordances between FT-AT and histology were attributable to biopsy failure (mostly due to small length) and 2% to FT-AT failure \[[@B9]\]. The present work allowed frequently asked questions to be answered, the first being whether the diagnostic values of FT-AT had been confirmed in all studies performed to date. A major strength of the studies pertaining to FT-AT is that they were carried out on a large number of patients with chronic hepatitis C, and the results were reproducible in different populations, including patients coinfected with HIV. There was a small variability in the AUROCs, both for the diagnosis of significant fibrosis (0.73 to 0.87) and significant necrosis (0.75 to 0.86). A weakness of this study was that the same group, which developed these tests, performed most of the published studies. However the independent published studies found the same significant diagnostic values than non-independent or multicentre studies. Several recent independent studies confirmed the predictive value of FT-AT \[[@B26],[@B30]\]. The second question concerned the comparison of FT-AT to other tests. In their recent review, Gebo et al. \[[@B10]\] concluded that panels of markers might have the greatest value in predicting the absence or no more than minimal fibrosis on biopsy, and in predicting the presence of cirrhosis on biopsy (Evidence Grade B). They pointed out that five studies \[[@B11],[@B32]-[@B35]\] used large panels of markers and achieved the greatest predictive values. Among these 5 studies were the first FT-AT study \[[@B11]\] and another study developed by the same group (combining age and platelets) \[[@B34]\]. A recent study compared FT-AT to the age and platelets index in the same patients and found that FT-AT was significantly better \[[@B15]\]. Three studies directly compared FT-AT, to hyaluronic acid \[[@B12]\], the Forns index \[[@B16]\] and the Wai index \[[@B17]\] in the same patients. FT-AT had higher diagnostic values (the AUROC was significantly higher). FT was in particular more sensitive for discriminating between F1 and F2, and more linearly correlated to stages when compared to those 3 other markers \[[@B12],[@B16],[@B17]\]. An additional weakness of the Forns index is the inclusion of cholesterol, which varies greatly in patients with genotype 3 \[[@B16]\]. The limitations of these three comparisons \[[@B12],[@B16],[@B17]\] are that they were retrospective and were performed by the same group. These comparisons, however, had no evident sources of bias. The comparison with the Forns Index \[[@B16]\] included all patients of the Imbert-Bismut et al.study (n = 323) \[[@B11]\], as the parameters belong to the routine biochemical tests. The comparison with the APRI index included 249/323 patients (77%) without any difference between included or non-included patients when all characteristics were compared \[[@B17]\]. The comparison with hyaluronic acid \[[@B12]\] included a total of 165 out of the 244 (68%) randomized patients pre-included. The 165 included patients did not differ from the 79 non-included patients according to the main characteristics. Among the 165 patients, the fibrosis index was assessed in 461 samples and hyaluronic acid in 457 samples \[[@B12]\]. Recently, a study using profiles of serum protein N-glycans found that a profile has a similar AUROC than FT for the diagnosis of compensated cirrhosis. When combined with FT this marker had 100% specificity and 75% sensitivity for the diagnosis of compensated cirrhosis, which is not significantly different from the 92% specificity and 67% sensitivity of the FT \[[@B26]\]. This study was independent and prospectively designed for taking FT as the comparison test. Only 24 patients with cirrhosis were included and no details were given concerning the causes of discordance between biopsy and biochemical markers. However FT-AT is the only panel of markers identified by an independent overview \[[@B9]\], which has been compared in the same patients with most of the other proposed markers. No studies were found that compared FT-AT with a panel of extra-cellular matrix markers \[[@B31]\]. Compared to other panels, FT-AT also allowed an estimation to be made not only of the fibrosis stage but also the necroinflammatory (histological) activity. The present analysis of the integrated database demonstrated that the diagnostic value of FT-AT did not depend on HCV genotype or viral load. However, because of the small number of patients included, studies in genotype 4, 5 and 6 would be useful. The present analysis also answered another frequently asked question concerning the predictive values for the intermediate stages of fibrosis. Contrary to the initial hypothesis, the diagnostic values of FT-AT for consecutive stages of fibrosis and grades of necroinflammatory activity were the same for both moderate and extreme stages and grades. Our interpretation is that the same overlap exists between all stages, which is mainly related to the sampling error of the biopsy. It is very reassuring that the medians of FT-AT are linearly associated with stages and grades (Figures [3A,3B](#F3){ref-type="fig"}). The linearity of this association became even more evident as a larger number of patients were included (data not shown). Finally, the integrated database allowed a simple conversion system to be proposed to clinicians between liver injury as estimated by the FT-AT and that as estimated by liver biopsy (Figure [4](#F4){ref-type="fig"}). One conventional way to express the diagnostic values of FT-AT was summarized using the cutoffs of the distribution by stages and grades (Tables [2](#T2){ref-type="table"} and [4](#T4){ref-type="table"}). The negative predictive value of FT for excluding significant fibrosis was excellent for the 0.31 cutoff (91%), as was the negative predictive value for excluding significant activity at the 0.36 cutoff of AT (85% negative predictive value). The positive predictive value of the 0.72 cutoff of FT for significant fibrosis was also high at 76%. This, however, may appear lower than the negative predictive value. There is a technical explanation owing to the prevalence of significant fibrosis, which was only 0.31 in this population. According to the excellent specificity (above 0.95), the positive predictive value increased rapidly in populations with more fibrosis (data not shown). We recently observed that the main reason for this was probably because most of the so-called false positives of the FT were in fact false negatives due to the small sampling size of liver biopsies \[[@B5],[@B9]\]. The same comments can be made concerning the positive predictive value of AT for significant necrosis with 77% at the 0.60 cutoff. Again, it is probable that a large proportion of so-called false positives of AT were in fact false negatives due to liver biopsies which were too small. The ideal study would be one using biopsies measuring 40 mm in length, as two samples of 20 mm each during laparoscopy. Only this very high quality biopsy can be considered as a true gold standard. Obviously this type of biopsy cannot be performed routinely as first line, but it could be recommended for clinical research. Conclusions =========== Based on these results, the use of the biochemical markers of liver fibrosis (FibroTest) and necrosis (ActiTest) can be recommended as an alternative to liver biopsy for the first line assessment of liver injury in patients with chronic hepatitis C. In clinical practice, liver biopsy should be recommended only as a second line test, i.e., in case of high risk of error of biochemical tests or in transplanted patients. For clinical research, only very high quality liver biopsy (as two samples of 20 mm each) can be considered as a gold standard for validation of new alternatives. Methods ======= Analysis of the literature -------------------------- We did a search for all publications and communications between February 2001 and March 2004 with the key words \"FibroTest\" and \"ActiTest\" in Medline and in the abstract books of hepatology, gastroenterology, internal medicine and infectious diseases annual meetings. Only publications or abstracts concerning FT-AT in chronic hepatitis C were included. Diagnostic value of FT-AT among published studies ------------------------------------------------- For each study we assessed the diagnostic value for the diagnosis of significant fibrosis (bridging fibrosis or stages F2, F3, F4 according to the METAVIR scoring system) and significant necroinflammatory activity (moderate or severe necrosis, grades A2 or A3 according to the METAVIR scoring system) by the area under the receiver operating characteristics curve (AUROC). For several databases it was possible to re-analyze the individual data and we looked at the sensitivity and specificity according to different thresholds (0.10, 0.30, 0.60 and 0.80). When FT-AT was compared to other biochemical tests, we also assessed the corresponding sensitivity and specificity according to several thresholds. Comparison of FT-AT diagnostic values with other biochemical markers -------------------------------------------------------------------- We selected studies using direct comparisons of diagnostic values in the same patients. The AUROCs were compared for the diagnosis of significant fibrosis (F2F3F4) and significant necrosis (A2A3). Integrated database ------------------- Patients were included in an integrated database if they belonged to a published population of patients with chronic hepatitis C. Liver biopsy was scored using the METAVIR scoring system and FT-AT was assessed using the recommended pre-analytical and analytical procedures \[[@B18],[@B20]\]. A published population of 300 prospectively analyzed blood donors was included as a control group \[[@B20]\]. Diagnostic value of FT-AT according to HCV genotype and viral load ------------------------------------------------------------------ Using the integrated database, we compared the AUROCs of FT-AT for the diagnosis of significant fibrosis (F2F3F4) and significant activity (A2A3) between 4 classes of genotype (1, 2, 3 and the rarer genotypes 4, 5, 6 grouped together). For viral load, only those assessed in the same laboratory were included in the comparison between AUROCs, and the median was used to define low and high viral loads (3,800,000 copies/ml) \[[@B8]\]. Diagnostic value of FT-AT according to stage and grade ------------------------------------------------------ Using the integrated database, we compared the diagnostic values according to different stages or grades. We compared the AUROCs for all possible combinations of stages and grades, including combinations with blood donors. This allowed, for example, a comparison to be made of the diagnostic value of FT for discriminating between F1 and F2 after excluding all other stages of the database. Liver biopsies -------------- In the integrated database, liver biopsies were processed using standard techniques. A pathologist who was unaware of the biochemical markers evaluated fibrosis stage and necrosis grade according to the METAVIR scoring system \[[@B22],[@B23]\]. Fibrosis was staged on a scale of 0 to 4: F0 = no fibrosis, F1 = portal fibrosis without septa, F2 = few septa, F3 = numerous septa without cirrhosis, F4 = cirrhosis. The grading of activity by the METAVIR system (based on the intensity of necroinflammatory activity, mainly on necrosis) was scored as follows: A0 = no necroinflammatory activity, A1 = mild activity, A2 = moderate activity, A3 = severe activity \[[@B22],[@B23]\]. Biochemical markers ------------------- We used the previously validated FT-AT \[[@B8],[@B9],[@B11]-[@B21]\]. FT-AT is a non-invasive blood test that combines the quantitative results of six serum biochemical markers \[alpha2-macroglobulin, haptoglobin, gamma glutamyl transpeptidase (GGT), total bilirubin, apolipoprotein A1 and alanine aminotransferase (ALT)\] with the patient\'s age and gender in a patented artificial intelligence algorithm (USPTO 6,631,330) to generate a measure of fibrosis stage and necroinflammatory grade in the liver. Statistical analysis -------------------- Corresponding stages and grades were calculated from median scores and 95% confidence intervals were observed in 1,270 patients and 300 healthy blood donors. The AUROC was used as a measure of discrimination, estimated using the empirical (non-parametric) method by DeLong et al. \[[@B36]\], and were compared using the paired method by Zhou et al. \[[@B36]\]. All analyses are performed on the NCSS software (Kaysville, Utah) \[[@B36]\]. Authors\' contributions ======================= TP and MM conceived the study, performed the statistical analysis, and wrote the manuscript. FIM, BH and DM carried out biochemical analyses. RP, DT, VR, and YB participated in the coordination of the study, and drafted the manuscript. AM participated in the design and coordination of assays in the control group. All authors read and approved the final manuscript. Acknowledgements ================ Thierry Poynard has grants from the Association pour la Recherche sur le Cancer (ARECA) and from the Association de Recherche sur les Maladies Virales Hépatiques	PubMed Central	2024-06-05T03:55:48.391074	2004-9-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522750/", "journal": "Comp Hepatol. 2004 Sep 23; 3:8", "authors": [ { "first": "Thierry", "last": "Poynard" }, { "first": "Françoise", "last": "Imbert-Bismut" }, { "first": "Mona", "last": "Munteanu" }, { "first": "Djamila", "last": "Messous" }, { "first": "Robert P", "last": "Myers" }, { "first": "Dominique", "last": "Thabut" }, { "first": "Vlad", "last": "Ratziu" }, { "first": "Anne", "last": "Mercadier" }, { "first": "Yves", "last": "Benhamou" }, { "first": "Bernard", "last": "Hainque" } ] }
PMC522751	Introduction ============ The rapid development of experimental tools like nuclear magnetic resonance (NMR), mass spectrometry (MS), tissue array analysis, phosphorylation of protein kinases, and fluorescence labeling combined with autoradiography on two-dimensional gels promises unprecedented, powerful strategies for the identification of the structure of metabolic and proteomic networks. What is common to these techniques is that they allow simultaneous measurements of multiple metabolites or proteins. At present, these types of measurements are in their infancy and typically limited to snapshots of many metabolites at one time point (e.g., with MS; \[[@B1],[@B2]\]), to short time series covering a modest number of metabolites or proteins (e.g., with NMR \[[@B3],[@B4]\], 2-d gels \[[@B5]\] or protein kinase phosphorylation \[[@B6]\]), or to tissue arrays \[[@B7]\] that permit the simultaneous high-throughput analysis of proteins in a single tissue section by means of antibody binding or MS. Nonetheless, it is merely a matter of time that these methods will be extended to relatively dense time series of many concentration or protein expression values. We will refer to these types of data as metabolic or proteomic profilesand to the time development of a single variable within such a composite profile as trace. The intriguing aspect of profiles is that they implicitly contain information about the dynamics and regulation of the pathway or network from which the data were obtained. The challenge for the mathematical modeler is thus to develop methods that extract this information and lead to insights about the underlying pathway or network. In simple cases, the extraction of information can be accomplished to some degree by direct observation and interpretation of the shape of profiles. For instance, assuming a pulse perturbation from a stable steady state, Vance et al.\[[@B8]\] present guidelines for how relationships between the perturbed variable and the remaining variables may be deduced from characteristics of the resulting time profiles. These characteristics include the direction and timing of extreme values (i.e., the maximum deviation from steady state) as well as the slopes of the traces at the initial phase of the response. Torralba et al.\[[@B9]\] recently demonstrated that these guidelines, applied to a relatively small set of experiments, were sufficient to identify the first steps of an in vitroglycolytic system. Similarly, by studying a large number of perturbations, Samoilov et al.\[[@B10]\] showed that it is possible to quantify time-lagged correlations between species and to use these to draw conclusions about the underlying network. For larger and more complex systems, simple inspection of peaks and initial slopes is not feasible. Instead, the extraction of information from profiles requires two components. One is of a mathematical nature and consists of the need for a model structure that is believed to have the capability of capturing the dynamics of the underlying network structure with sufficient accuracy. The second is computational and consists of fitting this model to the observed data. Given these two components along with profile data, the inference of a network is in principle a regression problem, where the aim is minimization of the distance between the model and the data. If a linear model is deemed appropriate for the given data, this process is indeed trivial, because it simply requires multivariate linear regression, which is straightforward even in high-dimensional cases. However, linear models are seldom valid as representations of biological data, and the alternative of a nonlinear model poses several taxing challenges. First, in contrast to linear models, there are infinite possibilities for nonlinear model structures. In specific cases, the subject area from which the data were obtained may suggest particular models, such as a logistic function for bacterial growth, but in a generic sense there are hardly any guidelines that would help with model selection. One strategy for tackling this problem is the use of canonical forms, which are nonlinear structures that conceptually resemble the unalterable linear systems models, but are nonlinear. Canonical models have in common that they always have the same mathematical structure, no matter what the application area is. They also have a number of desirable features, which include the ability to capture a wide variety of behaviors, minimal requirements for a prioriinformation, clearly defined relationships between network characteristics and parameters, and greatly enhanced facility for customized analysis. The best-known examples of nonlinear canonical forms are Lotka-Volterra models (LV; \[[@B11]\]), their generalizations \[[@B12]\], and power-law representations within the modeling framework of Biochemical Systems Theory (BST; \[[@B13]-[@B15]\]), most notably Generalized Mass Action (GMA) systems and S-systems. Lotka-Volterra models have their origin in ecology and focus strictly on interactions between two species at a time. Well-studied examples include competition processes between species, the dynamics of predators and prey, and the spread of endemic infections. In the present context it might seem reasonable to explore the feasibility of these models for the representation of the dynamics of proteins and transcription factor networks, but this has not been done so far. The strict focus on two-component interactions in LV models has substantial mathematical advantages, but it has proven less convenient for the representation of metabolic pathways, where individual reaction steps depend on the substrate, but not necessarily on the product of the reaction, or are affected by more than two variables. A simple example of the latter is a bi-substrate reaction that also depends on enzyme activity, a co-factor and possibly on inhibition or modulation by some other metabolite in the system. These types of processes have been modeled very successfully with GMA and S-systems. Between these two forms, the S-system representation has unique advantages for system identification from profiles, as was shown elsewhere \[[@B16]-[@B24]\] and will be discussed later in this article. In some sense, Karnaukhov and Karnaukhova \[[@B25]\] used a very simplified GMA system for biochemical system identification from dynamic data, in which all mono-substrate or bi-substrate reactions were of first order. This reduced the estimation to the optimization of rate constants, which the authors executed with an integral approach. The inference of a nonlinear model structure from experimental data is in principle a straightforward \"inverse problem\" that should be solvable with a regression method that minimizes the residual error between model and data. In practice, however, this process is everything but trivial (cf.\[[@B26]\]) as it almost always requires an iterative search algorithm with all its numerical challenges, such as the existence of multiple local minima and failure to converge. Recent attempts of ameliorating this problem have included Bayesian inference methods \[[@B27]\], similarity measures and correlation \[[@B28]\], mutual information \[[@B29]\], and genetic algorithms \[[@B30]\]. An indication of the complexity of nonlinear estimation tasks and their solutions is a recent pathway identification involving an S-system with five variables, which was based on a genetic algorithm \[[@B21]\]. The algorithm successfully estimated the parameter values, but although the system under study was relatively small and noise free, each loop in the algorithm took 10 hours on a cluster of 1,040 Pentium III processors (933 MHz). It is quite obvious that such an approach cannot be scaled up to systems of dozens or hundreds of variables. Nonlinear estimation methods have been studied for a long time, and while computational and algorithmic efficiency will continue to increase, the combinatorial explosion of the number of parameters in systems with increasingly more variables mandates that identification tasks be made easier if larger systems are to be identified. One important possibility, which we pursue here, is to prime the iterative search with high-quality starting conditions that are better than naïve defaults. Clearly, if it is possible to identify parameter guesses that are relatively close to the true, yet unknown solution, the algorithm is less likely to get trapped in suboptimal local minima. We are proposing here to obtain such initial guesses by preprocessing the temporal profile data and fitting them preliminarily by straightforward multivariate linear regression. The underlying assumption is that the structural and regulatory connectivity of the network will be reflected, at least qualitatively, in the regression coefficients. D\'haeseleer et al.\[[@B31]\] explored a similar approach for analyzing mRNA expression profiles, but could not validate their results because they lacked a mechanistic model of gene expression. Furthermore, because of the unique relationship between network structure and parameters in S-system models (see below), we will demonstrate that it is possible to translate the regression coefficients into constraints on the parameter values of an S-system model and thereby to reduce the parameter search space very dramatically. Several other groups have recently begun to target network identification tasks with rather diverse strategies. Chevalier et al.\[[@B32]\] and Diaz-Sierra and co-workers \[[@B33],[@B34]\] proposed an identification approach that is similar to the one proposed here in some aspects, though not in others. These authors also used linearization of a nonlinear model, but based their estimation on measured time developments of the system immediately in response to a small perturbation. These measurements were used to estimate the Jacobian of the system at the steady state. In contrast to this focus on a single point, we are here using smoothed long-term time profiles and do not necessarily require system operation at a steady state. Also using linearization, Gardner et al.\[[@B35]\] recently proposed a method of network identification by multiple regression. However, they only considered steady-state measurements as opposed to temporal profiles. It is known from theoretical analyses (e.g., \[[@B15],[@B36]\]) that different dynamical models may have the same steady state and that therefore steady-state information alone is not sufficient for the full characterization of a network. Mendes and Kell \[[@B37]\] used a neural network approach for an inverse problem in metabolic analysis, but their target system was very small and fully known in structure. Furthermore, their data consisted of a \"large number of steady-state simulations\", rather than the limited number of time traces on which our analysis is based. Chen et al.\[[@B38]\] used neural networks and cubic splines for smoothing data and identifying rate functions in otherwise linear mass-balance models. Methods ======= The behavior of a biochemical network with nspecies can often be represented by a system of nonlinear differential equations of the generic form ![](1742-4682-1-8-i1.gif) where Xis a vector of variables X~i~, i= 1, \..., n, fis a vector of nonlinear functions f~i~, and μis a set of parameters. If the mathematical structure of the functions f~i~is known, the identification of the network consists of the numerical estimation of μ. In addition to the challenges associated with nonlinear searches mentioned above, this estimation requires numerical integration of the differential equations in (1) at every step of the search. This is a costly process, requiring in excess of 95% of the total search time; if the differential equations are stiff, this percentage approaches 100% \[[@B39]\]. A simplification, which circumvents the problem of integration, consists of substituting the system of differential equations with decoupled algebraic equations by replacing the differentials on the left-hand side of Eq. (1) with estimated slopes \[[@B16],[@B17]\]. Thus, if the system consists of ndifferential equations, and if measurements are available at Ntime points, the decoupling leads to n× Nalgebraic equations of the form ![](1742-4682-1-8-i2.gif) It may be surprising at first that it is valid to decouple the tightly coupled system of nonlinear differential equations. Indeed, this is only justified for the purpose of parameter estimation, where the decoupled algebraic equations simply provide numerical values of variables (metabolites or proteins) and slopes at a finite set of discrete time points. The experimental measurements thus serve as the \"data points,\" while the parameters μ~ij~are the \"unknowns\" that need to be identified. The quality of this decoupling approach is largely dependent on an efficient and accurate estimation of slopes from the data. Since the data must be expected to contain noise, this estimation is a priorinot trivial. However, we have recently shown \[[@B23],[@B39]\] that excellent estimates can be obtained by smoothing the data with an artificial neural network and computing the slopes from the smoothed traces (see Appendix for detail). Different Linearization Approaches ---------------------------------- The smoothing and decoupling approach reduces the cost of finding a numerical solution of the estimation task considerably. Nonetheless, algorithmic issues associated with local minima and the lack of convergence persist and can only be ameliorated with good initial guesses. To this end, we linearize the model fin Eq. (1) about one or several reference states. As long as the system stays close to the given reference state(s), this linearization is a suitable and valid approximation. We consider four options: (I) linearization of absolute deviations from steady state; (II) linearization of relative deviations from steady state; (III) piecewise linearization; and (IV) Lotka-Volterra linearization. Option (I) is based on deviations of the type z~i~= X~i~- X~ir~, where X~ir~denotes the value at a reference state of choice. If the reference state is chosen at a stable steady state, the first-order Taylor-approximation is given by ![](1742-4682-1-8-i3.gif) where Ais the n × nJacobian with elements a~ij~= (df~i~/ dX~j~) calculated at X~r~(cf.\[[@B32]-[@B34]\]). If the reference state is not chosen at a steady state, the equation contains an additional constant term a~i0~, which is equal to f~i~(X~r~). For option II, we define a new variable u~i~= z~i~/X~ir~. At a steady state, this yields the linear system ![](1742-4682-1-8-i4.gif) where A\'is an n × nmatrix in which a\'~ij~= (X~jr~/ X~ir~)·a~ij~. A general concern regarding linearization procedures is the range of validity of sufficiently accurate representation, which is impossible to define generically. From an experimental point of view, the perturbations from steady state must be large enough to yield measurable responses. This may require that they be at the order of 10% or more. Depending on the nonlinearities in f, a perturbation of this magnitude may already lead to appreciable approximation errors. While this is a valid argument, it must be kept in mind that the purpose of this priming step is simply to detect the topological structure of connectivity and not necessarily to estimate precise values of interaction parameters. Simulations (see below) seem to indicate that this detection is indeed feasible in many cases, even if the deviations are relatively large. In order to overcome the limitation of small perturbations, a piecewise linear regression (option III) may be a superior alternative. In this case, we subdivide the dataset into appropriate time intervals and linearize the system around a chosen state within each subset. Most (or all) reference states are now different from the steady state, with the consequence that Eq. (3) has a constant term a~i0~, which is equal to f~i~(X~r~). The choice of subsets and operating points offers further options. In the analysis below, we use the locations of extreme values (maximum deviation from steady state) of the variables as the breakpoints between different subsets. Thus, a variable with a maximum and a later minimum has its time course divided into three subsets. The fourth alternative (option IV) is a Lotka-Volterra linearization. In a Lotka-Volterra model, the interaction between two species X~i~and X~j~is assumed to be proportional to the product X~i~X~j~\[[@B11]\]. Furthermore accounting for linear dependence on the variable of interest itself, the typical Lotka-Volterra equation for the rate of change in X~i~is ![](1742-4682-1-8-i5.gif) The right-hand side of this nonlinear differential equation becomes linear if both sides are divided by X~i~, which is usually valid in biochemical and proteomic systems, because all quantities of interest are non-zero. Thus, the differentials are again replaced by estimated slopes, the slopes are divided by the corresponding variable at each time point, and fitting the nonlinear LV model to the time profiles becomes a matter of linear regression that does not even require the choice of a reference state. The quality of this procedure is thus solely dependent on the quality of the data and ability of the LV model to capture the dynamics of the observed network. It is known (e.g., \[[@B11],[@B40]\]) that the mathematical structure of LV models is rich enough to model any nonlinearities, if sufficiently many equations are included. However, there is no general information about the quality of fit in particular modeling situations. Regression ---------- No matter which option is chosen, the next step of the analysis consists of subjecting all measured time traces to multivariate linear regression and solving for the regression coefficients (i.e., v~ij~\'s and w~i~\'s, or α~ij~\'s). The response variable is the rate of change of a metabolite, while the predictors are the concentrations of each metabolite in the network. The different linearization models (I-IV) differ in the transformations of the original datasets, which are summarized in Table [1](#T1){ref-type="table"}. For example, the response variable of the linear model in Eq. (4) is given by y~i~= ![](1742-4682-1-8-i6.gif) /X~ir~, and the predictor variables are transformed as x~i~= (X~i~- X~ir~)/X~ir~. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Transformation of data for regression analysis ::: RESPONSE VARIABLE PREDICTOR VARIABLE ---------------------------------------------- --------------------- ------------------------------- A. Absolute deviation from a reference state y~i~= x~i~= X~i~- X~ir~ B. Relative deviation from a reference state C. Lotka-Volterra system x~i~= X~i~ We assume the general linear model is y= a~i0~+ Σ(a~ij~x~j~). The X~i~denote experimental time series data for metabolite i, while the slopes () are estimated from the smooth output functions of the artificial neural network that had been trained on the experimental data. Subscript rdenotes the value of the metabolite at a reference state. Linearization options I and II are included in transformations A and B respectively, assuming that the reference state is a steady state. For a piecewise linear linearization (option III), the data may be transformed following either A or B. ::: The result of the regression is a matrix of coefficients that indicate to what degree a metabolite X~j~affects the dynamics (slope) of another metabolite X~i~. In particular, a coefficient that is zero or close to zero signals that there is no significant effect of X~j~on the slope of X~i~. By the same token, a coefficient that is significantly different from zero suggests the presence of an effect, and its value tends to reflect the strength and direction of the interaction. In either case, the coefficients computed from the linear regression provide valuable insight into the connectivity of the network. Furthermore, the estimated coefficients provide constraints on the parameter values of the desired nonlinear model f. Indeed, if fconsists of an S-system model, the coefficients estimated from the regression can be converted into combinations of S-system parameters, as is demonstrated in the following theoretical section and illustrated later with a specific example. Relationships between Estimated Regression Coefficients and S-system Parameters ------------------------------------------------------------------------------- The regression analysis yields coefficients that offer information on the connectivity of the network of interest. It also provides clues about the parameter values of the underlying nonlinear network model fin Eq. (1) if this model has the form of an S-system. To determine the relationships between the regression coefficients and the parameters of the S-system, it is convenient to work backwards by computing the different types of linearizations discussed before for the particular case of S-system models. This derivation is simply a matter of applying Taylor\'s theorem. In the S-system formalism, the rate of change in each pool (variable) is represented as the difference between influx into the pool and efflux out of the pool. Each term is approximated by a product of power-law functions, so that the generic form of any S-system model is ![](1742-4682-1-8-i10.gif) where nis the number of state variables \[[@B13],[@B14]\]. The exponents g~ij~and h~ij~are called kinetic ordersand describe the quantitative effect of X~j~on the production or degradation of X~i~, respectively. A kinetic order of zero implies that the corresponding variable X~j~does not have an effect on X~i~. If the kinetic order is positive, the effect is activating or augmenting, and if it is negative, the effect is inhibiting. The multipliers α~i~and β~i~are rate constantsthat quantify the turnover rate of the production or degradation, respectively. If the Taylor linearization is performed at a steady state, the production term of the S-system model equals the degradation term. The absolute deviation of the first option, z~i~= X~i~- X~is~, where the subscript sdenotes the value of the variable at steady state, then leads directly to ![](1742-4682-1-8-i11.gif) where c~ij~= g~ij~- h~ij~, ![](1742-4682-1-8-i12.gif) (cf. \[[@B41]\]). The so-called F-factors F~ij~are always non-negative, while c~ij~may be either positive or negative depending on the relationship between X~i~and X~j~. A common scenario is that a variable X~j~influences either the production or degradation of variable X~i~, but not both. In this case, a positive (negative) c~ij~implies activation (inhibition) of production or inhibition (activation) of degradation. The special case of c~ij~= 0 permits two possible interpretations: 1) g~ij~= h~ij~= 0, which implies that X~j~has no effect on either production or degradation of X~i~; or 2) g~ij~= h~ij~≠ 0, which means that X~j~has the same effect on both production and degradation of X~i~. The former case is the more likely, but there are examples where the latter may be true as well, and this is indeed the case in the small gene network in Figure [1](#F1){ref-type="fig"}. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Test System.a) Gene network \[42\] used as test system for illustrating the proposed methods. Solid arrows represent material flow, while dashed arrows indicate regulatory signals that either activate (+) or inhibit (-) a process. The network contains two genes, Gene 1 and 2. X~1~is the mRNA produced from gene 1, X~2~is the enzyme for which the gene codes, and X~3~is an inducer protein catalyzed by X~2~. X~4~is the mRNA produced from Gene 2 and X~5~is a regulator protein for which the gene codes. Positive feedback from X~3~and negative feedback from X~5~are assumed in the production of mRNAs from the two genes. b) S-system model of the gene network, according to Hlavacek and Savageau \[42\] and Kikuchi et al.\[21\]. ::: ![](1742-4682-1-8-1) ::: Comparing the expression in Eq. (6) with the linear regression results, one sees immediately that each coefficient a~ij~in Eq. (3) corresponds to the product of F~ij~and c~ij~: a~ij~= F~ij~c~ij~. (7) Thus, once the regression has been performed and the coefficients a~ij~have been estimated, the parameters of the corresponding S-system are constrained -- though not fully determined -- by Eq. (7). In particular, Eq. (7) does not allow a distinction between various combinations of g~ij~and h~ij~, as long as the two have the same difference. For instance, re-interpreting the regression coefficients as S-system parameters does not differentiate between the overall absence of effect of X~j~on X~i~(g~ij~= h~ij~= 0) and the same effect of X~j~on both the production and degradation of X~i~(g~ij~= h~ij~≠ 0). This observation is related to the observation of Sorribas and Cascante \[[@B36]\] that steady-state measurements are insufficient for completely identifying an S-system model. Relative deviations from steady state, u~i~= (X~i~- X~is~) / X~is~, in option II, are assessed in an analogous fashion. In this case one obtains ![](1742-4682-1-8-i13.gif) where c~ij~= g~ij~- h~ij~, ![](1742-4682-1-8-i14.gif) \[[@B41]\]. Again, the F-factors F~i~are positive, while c~ij~may be either positive or negative. The piecewise linear model for an S-system is easily derived as well. It is given as ![](1742-4682-1-8-i15.gif) where X~jr~denotes the value of the variable at the reference state. This case also includes the situation of a single approximation, which however is not necessarily based on a steady-state operating point. In the case of the Lotka-Volterra linearization, the correspondence between computed regression coefficients and S-system parameters is determined most easily by dividing the S-system equations by the corresponding X~i~and then linearizing around an operating point. The resulting expressions become especially simple if this point is chosen as the steady state. In this case, the relationship between the parameters of the LV system and the S-system are ![](1742-4682-1-8-i16.gif) where c~ij~= g~ij~- h~ij~. Results ======= We applied the methods described in the previous sections to simulated time profiles obtained from the small gene network in Figure [1a](#F1){ref-type="fig"}. Hlavacek and Savageau \[[@B42]\] modeled this network as an S-system with five differential equations (Figure [1b](#F1){ref-type="fig"}), and Kikuchi et al.\[[@B21]\] used it recently for exploring computational features of their proposed structure identification algorithm. The benefit of working with a known model is that complete information is available about both its structure and parameter values. In particular, it is possible to perform any number of experiments and to produce data and slopes with predetermined noise levels, which is not typically possible with real data. For this analysis, we thus used simulated noise free \"data,\" which allowed us to skip the neural network step of smoothing \[[@B23],[@B39]\]. To generate time profiles, the system was implemented with the parameter values published by Hlavacek and Savageau \[[@B42]\], and as in the analysis of Kikuchi et al.\[[@B21]\], the model was initialized with various perturbations from steady state and numerically integrated over a sufficient time horizon to allow the system to return to the steady state. Preliminary Analysis -------------------- Quasi as a pre-analysis, we examined the guidelines proposed by Vance et al.\[[@B8]\]. Indeed, the results show that many of these are applicable to the gene regulatory network. The order of the extrema (i.e., the maximum deviations from steady state) of the various variables both in time and size is in accordance with their \"topological distance\" from the perturbed variable, and variables not directly affected by the perturbed variable have zero initial slopes. As an example, the effect of a perturbation in X~3~is shown in Figure [2](#F2){ref-type="fig"}. All variables increase in response, with variables X~1~and X~4~reaching their maximal deviation from steady state before X~2~and X~5~, suggesting that X~1~and X~4~precede X~2~and X~5~in the pathway. The value of the initial slope is different from zero for X~1~and X~4~, implying that these variables are directly affected by X~3~, whereas X~2~and X~5~have zero initial slopes suggesting that their responses are mediated through other variables. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### *Dynamic response of the network after a perturbation in X~3~The response is shown as relative deviation from steady state. The guidelines proposed by Vance et al.\[8\] indicate that X~1~and X~4~precede X~2~and X~5~because they reach their maximum deviation earlier and the maximal values are larger than those of X~2~and X~5~. All variables respond in a positive manner, which implies either a mass transfer or positive modulation (activation). The system determined from this analysis is essentially the same as in Figure 1a. The only relationship missed is the effect of X~2~on the production and degradation of X~3~. ::: ![](1742-4682-1-8-2) ::: Maximal information about the network is obtained when every variable is perturbed sequentially. Experimentally, such perturbations could be implemented with modern methods of RNA interference \[[@B43]\] or, for biotechnological purposes, in a chemostat \[[@B9]\]. In our model case, we can actually identify all kinetic orders that are zero in the original model, and this amounts to determining the connectivity of the pathway. The only relationship this analysis does not pick up is the effect of X~2~on X~3~. This result is not surprising, because the effect of X~2~is the same on both the production and degradation of X~3~, which leads to cancellation. It is noted that this analysis does not necessarily distinguish between transfer of mass and a positive modulation, because both result in a positive effect on a variable. In a realistic situation, biological knowledge may exclude one of the two options, as in this case, where modulation is the only possibility for the effect of X~3~on both X~1~and X~4~, because the former is a protein and the latter are RNA transcripts. For the mathematical model in the S-system form, this is not an issue, as both types of influence are included in the equations in the same way (as a positive kinetic order). Regression Analysis ------------------- While Vance\'s method works well in this simple noise-free system, it is not scalable to larger and more complex systems. The next step of our analysis is therefore regression according to the four options presented above and with a number of simulated datasets of the gene network that differ in the variable to be perturbed and the size of the perturbation. Because the illustration here uses a known model and artificial data, it is easy to compute the true regression coefficients through differentiation of the S-system model. These coefficients can be used as a reference for comparisons with coefficients computed from the entire time traces, which mimics the estimation process for (smoothed) actual data. ### Options I, II and IV The results for three of the options (I, II and IV) can be summarized in the following three points, while the piecewise linear model will be discussed afterwards. \(1) The network connectivity is reflected in the values of the regression coefficients.The values of the estimated coefficients provide strong indication as to which variables have a significant influence on the dynamics of other variables. A comparison between computed and estimated coefficients is shown in Table [2](#T2){ref-type="table"} for the linear model with relative deviations (option II, Eq. 8). Most of the coefficients that in reality are zero (for example a~12~and a~24~) are not estimated as exactly zero, but their values are at least one order of magnitude smaller than the coefficients that are in actuality not zero. Table [2](#T2){ref-type="table"} also indicates that not all coefficients reflect the network correctly. The linear regression gives especially poor estimates for the coefficients associated with variables X~3~and X~4~. A possible explanation for X~3~is that the effect of X~2~is present in the non-linear system, but not in the linear system, and thus the behavior of X~3~must be explained by the other variables. Overall, of the 25 theoretically possible connections, 76% are correctly identified, while 24 % are false positives. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of computed and estimated coefficients ::: Computed coefficients* Estimated coefficients --------- --------------------------- ---------------------------- a10 0 0.0000 a11 -14.6780 -14.3647 a12 0 -0.1466 a13 7.3390 7.3414 a14 0 -0.2165 a15 -7.3390 -7.1723 a20 0 0.0000 a21 14.6780 14.6119 a22 -14.6780 -14.6540 a23 0 -0.0009 a24 0 0.0494 a25 0 -0.0309 a30 0 0.0000 a31 0 -2.3527 a32 0 1.3989 a33 -27.2517 -27.9204 a34 0 1.7491 a35 0 -0.9955 a40 0 0.0000 a41 0 2.0843 a42 0 -1.0925 a43 18.5664 19.0295 a44 -18.5664 -20.2112 a45 -9.2832 -8.3594 a50 0 0.0000 a51 0 -0.4026 a52 0 0.1384 a53 0 -0.0059 a54 18.5664 18.8987 a55 -18.5664 -18.7852 Regression coefficients for the small gene network (Figure 1), linearized about the steady state and based on relative deviations (option II). The first and second columns contain the computed and estimated regression coefficients, respectively. The regression coefficients a~ij~refer to the influence of variable jon variable i, while a~i0~is the constant term in each regression model. As the table indicates, the correspondence is good, except for the coefficients relating to X~3~and X~4~(see Text for explanation). The dataset consisted of 401 data points in the interval \[0,4\] and resulted from a simulation in which X~3~was perturbed at t= 0 to a value 5% above its steady-state value. ::: \(2) The different linear models give (qualitatively) the same results.A comparison of the results of the three models reveals that the values of the regression coefficients are very similar (see Table [3](#T3){ref-type="table"}). The same applies to their signs. Most important, all models correctly identify the connections present in the gene network. They also equally infer the same incorrect relationships. As an example, consider the coefficients associated with X~4~: all models infer the net positive effect of X~3~and the net negative effect of both X~4~and X~5~. At the same time, they also suggest that X~1~and X~2~have a significant effect on the dynamics of X~4~. In reality, they do not directly influence X~4~(see Figure [1](#F1){ref-type="fig"}), and it may be that their indirect effect, which is mediated by X~3~, is causing the false positive result. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Comparison of the different linearization options (I, II and IV) ::: I. Absolutedeviation II. Relative deviation IV. Lotka-Volterra --------- -------------------------- ---------------------------- ------------------------ a10 0.0000 0.0000 14.4748 a11 -14.3647 -14.3647 -18.9581 a12 -0.1466 -0.1466 -0.6836 a13 5.3878 7.3414 7.3367 a14 -0.1712 -0.2165 -0.4694 a15 -5.6702 -7.1723 -7.4981 a20 0.0000 0.0000 0.0144 a21 14.6119 14.6119 19.8910 a22 -14.6540 -14.6540 -19.9277 a23 -0.0006 -0.0009 -0.0001 a24 0.0390 0.0494 0.0472 a25 -0.0245 -0.0309 -0.0335 a30 0.0000 0.0000 26.4020 a31 -3.2058 -2.3527 2.8725 a32 1.9062 1.3989 -1.7989 a33 -27.9204 -27.9204 -26.6164 a34 1.8842 1.7491 -1.5871 a35 -1.0724 -0.9955 0.9692 a40 0.0000 0.0000 8.0270 a41 2.6365 2.0843 6.3364 a42 -1.3820 -1.0925 -4.1579 a43 17.6654 19.0295 19.0005 a44 -20.2112 -20.2112 -23.1319 a45 -8.3594 -8.3594 -7.7047 a50 0.0000 0.0000 0.0869 a51 -0.5092 -0.4026 -0.6617 a52 0.1751 0.1384 0.4441 a53 -0.0055 -0.0059 -0.0003 a54 18.8987 18.8987 20.2939 a55 -18.7852 -18.7852 -20.2152 Estimated coefficients for three of the linearization approaches: absolute deviation from steady state (left column), relative deviation from steady state (center column) and Lotka-Volterra linearization (right column). The dataset consisted of 401 data points in the interval \[0,4\] and resulted from a simulation in which X~3~was perturbed at t= 0 to a value 5% above its steady-state value. ::: \(3) The greater the perturbation, the less accurate is the estimation of the regression coefficients.The deviation between the estimated and computed coefficients increases as the size of the perturbation increases (see Table [4](#T4){ref-type="table"}). For the models obtained by linearizing about the steady state (Eqs. (6) and (8)), this is an expected result, as the Taylor-expansion only gives a valid approximation close to steady state. For these systems, \"close\" may correspond to a perturbation of less than 5--10% with respect to the steady-state value. Nonetheless, the greater perturbations still give a relatively good picture in terms of the connectivity of the system. For a 5% perturbation, the fraction of correctly identified connections is 76% and for a two-fold perturbation it is still 64 %. Perturbations of more than 5--10 % of the steady state also cause problems for the Lotka-Volterra model, from which one might have expected a higher tolerance as the linearization is independent of a reference state. It seems that the dynamics of the true system in our particular example is about equally well modeled by the nonlinear LV-model as by the linear models. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### The effect of the size of the perturbation ::: Computed 5 % 10 % 50 % 200 % --------- -------------- ---------- ---------- ---------- ----------- a10 0 0.0000 0.0000 0.0001 0.0008 a11 -14.6780 -14.3647 -14.1817 -13.1496 -11.3439 a12 0 -0.1466 -0.1429 -0.0671 0.5735 a13 7.3390 7.3414 7.3438 7.3598 7.3735 a14 0 -0.2165 -0.3673 -1.2462 -2.7619 a15 -7.3390 -7.1723 -7.0780 -6.4846 -5.2501 a20 0 0.0000 0.0000 0.0000 -0.0003 a21 14.6780 14.6119 14.5748 14.4207 14.5029 a22 -14.6780 -14.6540 -14.6623 -14.7503 -15.1862 a23 0 -0.0009 -0.0016 -0.0054 -0.0070 a24 0 0.0494 0.0839 0.2494 0.3462 a25 0 -0.0309 -0.0464 -0.1119 -0.0951 a30 0 0.0000 0.0000 0.0004 0.0038 a31 0 -2.3527 -4.5412 -18.2307 -46.8953 a32 0 1.3989 2.6336 9.8422 24.4004 a33 -27.2517 -27.9204 -28.5955 -34.0204 -54.4047 a34 0 1.7491 3.4009 14.0961 39.3252 a35 0 -0.9955 -1.8949 -7.0627 -15.4759 a40 0 0.0000 0.0000 -0.0001 0.0001 a41 0 2.0843 3.7814 14.7316 41.5863 a42 0 -1.0925 -1.7693 -5.5766 -13.2688 a43 18.5664 19.0295 19.4964 23.2397 37.1866 a44 -18.5664 -20.2112 -21.6608 -31.4631 -58.1065 a45 -9.2832 -8.3594 -7.6404 -3.2226 6.5808 a50 0 0.0000 0.0000 -0.0001 -0.0015 a51 0 -0.4026 -0.6581 -2.5848 -10.1097 a52 0 0.1384 0.0830 -0.1317 0.1582 a53 0 -0.0059 -0.0110 -0.0435 -0.0879 a54 18.5664 18.8987 19.1602 21.0620 27.2722 a55 -18.5664 -18.7852 -18.9201 -20.0013 -24.0836 Overall, the estimated coefficients deviate more strongly from the corresponding computed values as the perturbation increases. However, there are substantial differences between variables. The coefficients associated with variable X~2~, for example, are hardly influenced, while the coefficients associated with X~3~are strongly affected. Overall, the method seems to produce the best results for perturbation up to 10%. The datasets for the regression consisted of 401 data points in the interval \[0,4\] and the method of linearization was option II. ::: ### Option III The piecewise linear model was obtained by dividing the whole dataset into three smaller subsets for each variable. The first interval contained the data points from t= 0 to the time of the first extreme value for a given variable (in this case a maximum for all variables). For the perturbed variable (having its first extreme value at t= 0) the first limit point was given by the smallest of the limit points of the other variables. The second interval contained the data points from the first to the second extreme value (a minimum), while the third interval included the remaining data points. The midpoint of each interval was taken to be the reference state. The result of the piecewise linear regression for a 5% deviation in X~3~is given in Table [5](#T5){ref-type="table"}. The first subset does not reflect the interactions of the system especially well, whereas the other two subsets correctly classify 88% and 96%, respectively, of the true connections in the network. It is worth noting that the coefficients associated with X~3~in the two last subsets reflect the variable\'s connectivity to a much greater extent than the other linearization approaches. As the reference state is different from the steady state, the effect of X~2~is present in the linear system as well, and thus there is no compensation through the other variables. Another benefit is that the piecewise model tolerates larger perturbations. Even for a two-fold perturbation, the fraction of correctly identified coefficients in the last subset is 84%. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Results for piecewise linear regression ::: Interval 1 Interval 2 Interval 3 --------- ---------------- ---------------- ---------------- a10 0.1315 -0.0419 0.0000 a11 -42.3980 -14.1738 -14.5490 a12 0.0000 -0.8010 -0.0464 a13 8.9105 7.3653 7.6299 a14 12.7757 -0.3340 -0.1386 a15 -3.3476 -6.9121 -7.2940 a20 0.0567 -0.0197 0.0000 a21 -1.1939 14.4913 14.6792 a22 -32.3300 -14.5116 -14.6784 a23 0.6133 0.0057 -0.0205 a24 7.0917 0.1016 -0.0018 a25 7.9313 -0.1047 0.0067 a30 -0.7858 -0.0181 0.0000 a31 -130.3724 -0.2358 0.0021 a32 0.0000 0.3616 -0.0007 a33 -20.7724 -27.6129 -27.2551 a34 62.1525 0.3496 -0.0027 a35 19.1470 -0.1984 0.0006 a40 0.3164 -0.0709 0.0000 a41 -13.6819 1.1412 -0.0115 a42 0.0000 -2.1478 0.0015 a43 19.8295 18.8534 18.6927 a44 -13.3654 -19.5811 -18.5494 a45 -7.2135 -8.0985 -9.2792 a50 0.1617 -0.0393 0.0000 a51 -149.5199 -0.8195 0.0250 a52 -160.3341 0.8175 -0.0074 a53 5.7537 0.0580 -0.0304 a54 85.3050 19.0394 18.5356 a55 53.9745 -19.1183 -18.5623 The complete dataset is divided into three subsets for each variable, where the first and second extreme values serve as breakpoints. The datasets for the regression consisted of 401 data points in the interval \[0,4\] and resulted from a simulation in which X~3~was perturbed at t= 0 to a value 5% above its steady-state value. ::: ### Degree of Similarity as a Measure of Reliability If we compare the results of all four linearized models, the degree of similarity may provide a measure of how reliable the estimated coefficients are, assuming that an interaction identified in all models is more reliable than an interaction identified in only one or few of the models. Considering the piecewise linear model as three models, yielding a total of 6 models from one dataset, one may thus determine the most likely connectivity for the small gene network. The result is presented in Table [6](#T6){ref-type="table"}. Of the 25 possible connections, 12 were identified correctly in all models, either as being positive, negative or non-existent, while an additional 6 connections were correctly identified in either 4 or 5 of the six models. For these six, one of the models misidentifying the type of connection was the first subset of the piecewise linear approximation, which does not reflect the connectivity of the network especially well, as was shown in Table [5](#T5){ref-type="table"}. It is also worth noting that only one of the interactions associated with X~3~is identified correctly from comparing the six models. The classification of the remaining four connections varies greatly among the different models, and it is therefore impossible to deduce a type of interaction with sufficient reliability. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Collective inference of the gene network based on results from all linearizations ::: X1 X2 X3 X4 X5 ------ --------------- --------------- --------------- --------------- --------------- X1 - (100 %) 0 (67 %) + (100 %) 0 (83 %) - (100 %) X2 + (100 %) - (100 %) 0 (100 %) 0 (83 %) 0 (83 %) X3 ? ? - (100 %) ? ? X4 \+ (67 %) \- (67 %) + (100 %) - (100 %) - (100%) X5 \- (83 %) 0 (83 %) 0 (83 %) + (100 %) - (100 %) Each minus sign implies a negative influence; a plus sign implies a positive influence, while zero implies no influence. Bold symbols denote correctly identified interactions, and numbers in parentheses give the fraction of models that suggested positive identification. Question marks imply that no type of interaction was identified in more than 50% of the models. ::: ### Constraining the Parameter Values In addition to reflecting the connectivity, the coefficients provide likely parameter ranges or likely constraints on parameter values of the true model. As an example, consider variable X~1~. Table [6](#T6){ref-type="table"} indicates that the variables having a significant effect are X~1~, X~3~and X~5~. If so, the linear model in Eq. (8) suggests the following: ![](1742-4682-1-8-i17.gif) where ![](1742-4682-1-8-i18.gif) and the regression coefficients (a~ij~) are taken from the model in Eq. (4). The values of the variables at steady state are known. Because the kinetic orders may be positive or negative and the c~ij~may result from different combinations of g~ij~\'s and h~ij~\'s, it is not possible to deduce directly which exponent is greater than the other. However, in many cases one may have additional information on the system, which further limits the degrees of freedom (e.g., \[[@B23]\]). In addition, the steady-state equation ![](1742-4682-1-8-i19.gif) must be satisfied and provides yet another constraint. Discussion ========== Identifying the structure of metabolic or proteomic networks from time series is a task that most likely will require large, parallelized computational effort. The search space for the algorithms is typically of high dimension and unknown structure and very often contains numerous local minima. This generic and frequent problem may be ameliorated if the search algorithm is provided with good initial guesses and/or constraints on admissible parameter values. Here, we have shown that linear regression may provide such information directly from the types of data to be expected from future experiments. For illustrative purposes, we used artificial data from a known network, but all methods are directly applicable to actual profile data and scaleable to large systems. The coefficients estimated from the different regressions reflect the effect of one variable on another surprisingly well and thus provide a simple fashion of prescreening the connectivity of the network. In addition, the estimated coefficients provide constraints on the parameter values, if the alleged nonlinear model has the form of an S-system. To explore the pre-assessment of data as fully as feasible, we studied four linearization strategies: using an absolute deviation from steady state; a relative deviation from steady state; piecewise linearization; and Lotka-Volterra linearization. Interestingly, all models gave qualitatively similar results for the analyzed example, and this degree of similarity may provide a measure of how reliable the identified connections are. Specifically, of the 25 possible connections in the small gene network studied, 19 were identified correctly in at least 83 % of the regression analyses. A concern of any linearization approach is the validity of the linear approximation. However, as long as the perturbation from steady state remains relatively small, the estimated linear model is likely to be a good fit of the actual nonlinear model, at least qualitatively. This limitation may furthermore be alleviated by fitting the profile data in a piecewise linear fashion. As most reference states in this case are different from the steady state, this strategy has the added benefit that more of the true relationships within the nonlinear model are likely to be preserved. As an alternative, one could explore the performance of the so-called \"log-linear\" model, which is linear in log-transformed variables \[[@B44]\]. The Lotka-Volterra linearization did not perform as well as expected with regard to large perturbations. This may be a consequence of the particular example, which was originally in S-system form rather than in a form more conducive to the LV structure, which emphasizes interactions between pairs of variables. Since it is easy to perform the LV analysis along with the other regressions discussed here, it may be advisable to execute all four analyses. The illustrative model used for testing the procedure consisted of a relatively small system with only five variables and relatively few interactions. Nonetheless, one should recall that this very system required substantial identification time in a direct estimation approach \[[@B21]\]. In order to check how scaleable the results of the proposed linearization method are, the method should be tested on larger systems. Some preliminary analyses suggest that the method works well, but that the likelihood of misidentified connections may grow with the size of the system, as one might expect. At the same time, experience with actual biological networks, for instance in ecology and metabolism, suggests that larger systems are often more robust in a sense that they do not deviate as much from the steady state as smaller systems. If this trend holds in general, the linearization becomes a more accurate representation as larger networks are being investigated and the proposed methods will therefore yield more reliable initial indicators of network connectivity. Independent of these issues, the methods proposed here will very likely be more valuable for bigger systems than other methods that are presently available, because without some preprocessing of the data and effectively priming the search, as it is proposed here, the combinatorial explosion will most certainly gain the upper hand eventually. Competing interests =================== None declared. Authors\' contributions ======================= SRV performed the analysis and prepared the results. JS developed and implemented the neural network for computation of slopes. EOV developed the basic ideas and directed the project. Appendix ======== It was recently shown that good parameter estimates of S-system models from metabolic profiles might be obtained by training an artificial neural network (ANN) directly with the experimental data. The result of this training is a so-called universal functionwhich smoothes the data with predetermined precision and also allows the straightforward computation of slopes that can be used for network identification purposes. This appendix briefly outlines the procedure; details can be found in Almeida \[[@B45]\] and Voit and Almeida \[[@B24]\]. The ANN consists of three layers; one input layer, one hidden layer and one output layer. The input layer consists of the measurement time points, the hidden layer has no direct biological interpretation, and the output layer contains the metabolite concentrations or levels of protein expression that the ANN is being trained to represent. The node values of the ANN in the hidden layer are calculated from a linear combination of input values with different weights according to a multivariate logistic equation. Similarly, the values of the output layer are determined from linear combinations of the hidden node values with different weights, again using a multivariate logistic function. It is known that this type of nested multivariate logistic function has unlimited flexibility in modeling nonlinearities \[[@B46]\]. Noise and sample size do not have a devastating effect on the results of the ANN-method, as long as the true trend is well represented \[[@B39]\]. In fact, the ANN approach provides an unlimited number of sampling points, as values at any desired time points may be estimated from the universal output function. Finally, the calculation of the slopes of the smooth output functions is mathematically unwieldy, but computationally straightforward. The use of the entire time course is in stark contrast to earlier methods of parameter estimation and structure identification in metabolic networks. Mendes and Kell \[[@B37]\] applied their ANN-based parameter estimation to steady-state data, while we are using time profiles. Chevalier and co-workers \[[@B32]\] first fitted the nonlinear solution with a linear model (as shown in Eq. 3), expressed this solution in terms of eigenvectors and eigenvalues, and then obtained the slopes by numerical differentiation. Sorribas et al.\[[@B47]\] suggested a variation on this approach, based on discretizing the solution of Eq. (3) as z(t~k+\ 1~) = z(t~k~)exp(h·A), (A1) where his the step size. The problem is thereby reduced to a mulitilinear regression in which the matrix Φ= exp(h·A) is the output. Instead of estimating the slopes, they obtain the Jacobian directly by ![](1742-4682-1-8-i20.gif) expanded in its Taylor-series. This approach yields a faster convergence to the elements of the Jacobian than the one suggested by Chevalier et al. \[[@B32]\], but the regression of Eq. (A1) is very sensitive to noise and missing data points. Our approach takes advantage of the entire time course and is therefore less sensitive to the particularities of assessing a system at a single point. The ANN itself does not provide much insight, because it is strictly a black-box model, but it is a valuable tool for controlling problems that are germane to any data analysis, namely noise, measurement inaccuracies, and missing data. Acknowledgments =============== This research was carried out during S.R.V.\'s scientific visit at the Medical University of South Carolina. The work was supported by a Quantitative Systems Biotechnology grant (BES-0120288; E.O. Voit, PI) from the National Science Foundation, a National Heart, Lung and Blood Institute Proteomics Initiative through contract N01-HV-28181 (D. Knapp, PI), and an Interdisciplinary USC/MUSC grant (E.P. Gatzke, PI). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the sponsoring institutions.	PubMed Central	2024-06-05T03:55:48.396695	2004-9-14	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522751/", "journal": "Theor Biol Med Model. 2004 Sep 14; 1:8", "authors": [ { "first": "Siren R", "last": "Veflingstad" }, { "first": "Jonas", "last": "Almeida" }, { "first": "Eberhard O", "last": "Voit" } ] }
PMC522803	Background ========== Hepatosplenomegaly is a widespread but neglected condition affecting many communities in sub-Saharan Africa where both Schistosoma mansoniand malaria infections are endemic. Severely enlarged and hard organs may be an indicator of increased portal pressure, as well as pointing to an increased risk of portal hypertension and its sequelae, including oesophageal varices and haematemesis \[[@B1]\]. Historically, hepatomegaly in sub-Saharan Africa has been associated with S. mansoniinfection, and on a global scale up to 8.5 million individuals may be affected \[[@B2]\], with the prevalence of organomegaly generally highest in children \[[@B3]\]. Enlargement of spleens has often been associated with the level of malaria transmission \[[@B4],[@B5]\]. One prevailing suggestion has been that hepatosplenomegaly amongst younger children is attributable to malaria infection, whereas older children have acquired some level of immunity to malaria and hence if they do have hepatosplenomegaly it is likely to be attributable to schistosomiasis \[[@B6]\]. However, chronic splenomegaly associated with portal hypertension has been reported in Kenyan hospitals amongst residents from areas where both S. mansoniand malaria infections are endemic \[[@B7],[@B8]\], and it has long been suggested that the presence of both infections may confound attempts to quantify the impact of either one \[[@B9],[@B10]\]. Recently, it was observed that IgG3 antibody responses to P. falciparumschizont antigen (Pfs Ag) were higher in Kenyan children with S. mansoniinfection and hepatosplenomegaly compared with children with infection but no hepatosplenomegaly \[[@B11]\], further implicating exposure to malaria in childhood as a risk factor for severe hepatosplenic morbidity. Given that malaria and schistosome infections are often endemic in the same communities, and are both targets of large-scale but independent, intervention programmes, it is important to understand how continued exposure to one parasite may affect the outcome of intervention against the other. The current longitudinal study evaluated praziquantel treatment on the regression of hepatosplenomegaly amongst children living in an area with endemic S. mansoniinfection and seasonal malaria infection \[[@B12],[@B13]\]. Before treatment, there was a strong correlation between S. mansoniegg count and the degree of both hepatomegaly and splenomegaly \[[@B12]\]. However, spatial analysis then revealed heterogeneity in exposure to both infections, and in particular there was an exacerbation of splenomegaly amongst children living in an area of relatively high exposure to both malaria and S. mansoniinfections \[[@B14]\]. The prevalence of hepatomegaly and splenomegaly decreased slowly during the three-year follow-up period, but had not completely abated by the end of the study \[[@B13]\]. Here, we report that the level of exposure to malaria infection was likely to have been an important factor in limiting the resolution of hepatosplenomegaly in these children. Methods ======= Study area ---------- Mbeetwani is situated in the district of Makueni, half way between Nairobi and Mombasa and approximately 10 km east of the main highway. The ethnic background of the population is predominantly Akamba. During the dry seasons, the river Kambu is the main source of water. Running from west to east, the river lies to the north of the community (Figure [1](#F1){ref-type="fig"}). Surface water is present at the western end during the dry season, whereas the eastern end is dry and water is retrieved by digging wells into the riverbed. River water is drawn for drinking, cooking and other domestic purposes. Washing of clothes and bathing takes place at numerous sites, and animals are taken to the river to drink. Clinical examination -------------------- The cohort and examination procedure are described in detail elsewhere \[[@B12]\]. Briefly, 96 children aged between 6 and 16 were selected for ultrasound detectable hepatomegaly after a community survey. They were examined before treatment with praziquantel at 40 mg/kg, and then examined again in the same month in the following three years. Treatment was given again two years into the study. All surveys took place in October of each year, towards the end of the dry season. A total of 67 children attended all examinations, whereas 71 children attended at least the baseline and third year follow-up surveys. Clinical examination consisted of liver and spleen palpation whilst children were in a supine position. Un-palpable spleens and livers were recorded as such, otherwise extensions in cm below the rib cage along the mid-clavicular line (MCL) and mid sternal line (MSL) were measured for the liver, and extension below the rib cage along the mid-clavicular and mid axillary lines (MAL) was recorded for the spleen. The firmness of each palpable organ was recorded as either soft, firm, or hard, as described previously \[[@B12]\]. An accompanying ultrasound examination was conducted according to World Health Organisation guidelines \[[@B15]\] to test for the presence of periportal fibrosis. Both the clinicians and the ultrasonographers were blinded as to the location of children\'s domiciles and previous examination results, and the same team performed the examinations throughout the study period. Mapping ------- Details of the mapping procedure are given elsewhere \[[@B14]\]. Briefly, the longitude and latitude of each house were recorded using a Magellan GPS 315 receiver. The course of the river Kambu was determined by taking readings approximately every 100 metres along a ten kilometre stretch that extended beyond the borders of the study area. The approximately rectangular study area was divided into 4 sectors with similar numbers of houses by bisecting the north-south and east-west transects at 4 km east of the western boundary and 1 km south of the river. Assays ------ Since the surveys were undertaken during the dry season, parasitaemia was not expected to provide a good enough estimate of exposure. Therefore, anti-Pfs IgG3 responses were used as a proxy measure of recent exposure \[[@B16]\], measured by ELISA as described previously \[[@B17]\]. Parasitology ------------ For the quantification of S. mansonieggs, each child provided 3 stool samples before the baseline survey in October 1999. Two 50 mg Kato slides were prepared from each stool sample using the Kato-Katz procedure \[[@B18]\]. All slides were prepared and examined by the same team. Finger-prick blood samples were taken on the same day that clinical and ultrasound examinations were performed, for detection of malaria parasitaemia. Mollusciciding -------------- Regular mollusciciding of the river Kambu was undertaken with Bayluscide ^®^(Bayer CropScience AG, Alfred-Nobel-Str 50, D40789 Monheim am Rhein, Germany). Area-wide application and focal spraying using backpacks was undertaken twice yearly to coincide with periods when snail populations were reduced due to environmental conditions. Snail sampling activities at 9 sites on a monthly basis were undertaken to confirm that snail numbers were kept at very low levels (HC Kariuki, unpublished observations). Treatment and ethical considerations ------------------------------------ The objectives of the study were carefully explained to the local community including the parents and teachers before the start of any activities. Informed assent was obtained from the parents or guardians of the children. After the clinical and ultrasound examination in October 1999, all children were treated with a single dose of praziquantel (Distocide^®^, Shin Poon Pharmaceuticals, Seoul, Republic of Korea), at 40 mg/kg body weight. The children were again treated in 2001 and 2002 under the same regimen. No other source of treatment was available locally during this period. Tablets were administered together with a piece of bread and a soft drink in order to minimise gastrointestinal side effects. The study was approved by the Kenya Medical Research Institute Ethical Review Committee. Analysis -------- Baseline data for each parameter were compared with 3 year follow up data amongst 71 children who attended the first and last surveys to estimate the number of children within each sector who presented with a lower degree of organomegaly at the end of the study. Repeated-measures ANOVA of each organ enlargement parameter was also undertaken for the 67 children who attended all surveys, with age, sex and sector of residence as fixed factors. Since ANOVA does not indicate the direction of any change, a post-hocanalysis was undertaken within each sector to determine whether there was an overall increase or decrease in the extent of organomegaly. This was achieved by comparing baseline enlargement and enlargement after 3 years follow up for each parameter, using the Wilcoxon Signed Ranks test. All analyses were conducted in SPSS v11 (SPSS inc, 2001, Chicago). Results ======= Geographical distribution of exposure ------------------------------------- Figure [1](#F1){ref-type="fig"} illustrates the geographical distribution of houses within the study area, as well as the course of the river Kambu. A total of 70 houses were geo-referenced. Also given are the median and 25--75%ile range of egg counts and anti-Pfs IgG3 responses within each sector before the intervention. Details of the analysis of S. mansoniegg counts and IgG3 responses within each sector are given elsewhere \[[@B14]\]. Briefly, there was a significant clustering of high egg counts in children living at the western end of the study area (Sector A and Sector C). Significant clustering of the highest anti-Pfs IgG3 responses was observed amongst children living in the two northern sectors (Sector A and Sector B). Thus, sector A was an area of relatively high exposure to both infections, sector B was an area of relatively high exposure to malaria only, sector C was an area of relatively high exposure to S. mansonionly, and sector D was an area of low exposure to both infections. Temporal changes in organ consistency by sector ----------------------------------------------- Significant variation in the prevalence of hard spleens over time was observed within each sector (Table [1](#T1){ref-type="table"}), attributable to an overall decline in the number of children with hard and enlarged spleens (Figure [2a](#F2){ref-type="fig"}). The prevalence of hardened spleens was highest at all time points in the sector with relatively high anti-Pfs IgG3 responses and high egg counts (Sector A). By two years post-treatment, no child resident more than one kilometre from the river presented with a hard spleen. The prevalence of firm livers in all sectors varied over time, except in sector B (Table [1](#T1){ref-type="table"}). In all other sectors there was a decrease in the prevalence of firm livers over the course of the follow-up period (Figure [2b](#F2){ref-type="fig"}). At the end of the study, the lowest prevalence of firm livers was observed in children from sector D. Temporal changes in organomegaly by sector ------------------------------------------ Significant variation in the degree of MAL splenomegaly was observed in each sector over the three-year follow-up (F = 20.7, p \< 0.001). Figure [3](#F3){ref-type="fig"} illustrates a decline in the degree of MAL splenomegaly within each sector, and this was confirmed by post-hocWilcoxon Rank analysis (Table [2](#T2){ref-type="table"}). The rate of regression was affected independently neither by sector, age nor sex, and children from sector A presented with the largest spleens at each time point. At the third year post treatment follow-up, the fraction of each group presenting with no MAL splenomegaly in each sector was 10/24 (Sector A), 6/14 (Sector B), 13/15 (Sector C), and 16/20 (Sector D). MCL splenomegaly also varied significantly over time (Figure [3b](#F3){ref-type="fig"}, F = 14.1, p \< 0.001), but the extent of variation was affected by sector (F = 2.191, p = 0.024). Post-hocWilcoxon rank analysis revealed that there was a significant decrease in the extent of MCL splenomegaly only in sector D (Table [2](#T2){ref-type="table"}). MSL liver enlargement varied significantly over time (F = 3.78, p = 0.012), with the variation being affected by sector (F = 2.57, p = 0.008). There was no consistent decline in MSL hepatomegaly over time within any one sector (Figure [3c](#F3){ref-type="fig"}). Although a fraction of the children in each sector presented with reduced hepatomegaly at the third year follow-up, Wilcoxon Rank analysis of the direction of change within each sector confirmed that an overall significant decrease in the extent of MSL hepatomegaly between the baseline survey and the third year follow-up occurred only amongst children from sector D (Table [2](#T2){ref-type="table"}). Children from sector B presented with the least improvement in MSL hepatomegaly, with 10/12 exhibiting no change or an increase in this parameter when baseline data were compared with data from the third year follow-up. MCL hepatomegaly values could not be used in the repeated measures analysis due to a lack of normal variation in the data. Wilcoxon rank analysis within sectors revealed no significant change in this measurement between baseline and the third year follow-up in any sector (Table [2](#T2){ref-type="table"}). Ultrasound results ------------------ There was no evidence of periportal fibrosis in any of the children at any time point. Discussion ========== Many clinical surveys of sub-Saharan communities have reported that organomegaly of the spleen or liver is a very common condition, but attribution to a specific aetiology has always been problematic. Distinguishing the contribution of malaria from that of schistosome infections is particularly complex since they are often endemic in the same communities. However, it is well established that exposure to both S. mansoniand malaria varies over small areas \[[@B19]-[@B21]\], and the potential therefore exists to compare morbidity in areas of overlapping exposure with areas where infections of one species are more prevalent. Here, we have exploited observations concerning micro-geographical variation in the distribution of each infection to demonstrate how the benefits of treatment with praziquantel on hepatosplenomegaly may be affected by local heterogeneity in malaria exposure. Our results further implicate both parasites as aetiological agents of chronic hepatosplenomegaly including firmness of the organs in school-aged children, and have strong implications not only for estimating the burden of S. mansoniand malaria, but also for estimating the outcome of interventions. The generation of these results was facilitated by several important features of the study design. The intermittent mollusciciding of the river Kambu was particularly important, since it reduced the potential for morbidity to rebound as a result of re-infection, as has been seen elsewhere in studies of both S. mansoniand S. haematobium\[[@B22]\]. The timing of the follow-up surveys outside the malaria transmission season was another important component, since it allowed an assessment of the clinical situation without the confounding effects of transient morbidity associated with acute malaria. Without these particular features, it is unlikely that we could have come to any conclusions about the effects of chronic exposure to malaria infection on the regression of hepatosplenomegaly. Another important component of the study was the mapping of both house co-ordinates and the local river. By doing so, we were able to identify spatial clustering of relatively high S. mansoniegg counts at the western end of the study area. Such clustering is likely related to variation in the amount of surface water along the river, and hence variation in transmission potential due to the strong relationship between snail abundance and water level observed during long term studies of transmission in this area \[[@B23]\]. The clustering of anti-Pfs IgG3 responses to schizont antigen along a tract parallel to the course of the river suggests a sharp decrease in transmission further from the river \[[@B14]\]. Although we observed several significant changes following praziquantel treatment of the cohort, it is important to note that this was a retrospective analysis, which carries a few limitations. The observations made in this study were based on a small, case-only cohort, and therefore do not represent the outcome of an intervention programme involving mass treatment of a population. As this was a retrospective analysis, no control was possible concerning treatment for malaria during the follow-up period. The population had access to antimalarial drugs at local shops, but so far as we are aware there was no systematic intervention against malaria infection during the follow-up period. The results therefore encompass the effects of background treatment for, as well as exposure to, malaria infections. Reliability of the clinical measurements is an important factor, especially given the size of the cohort. We have assessed this procedure elsewhere and it has been found to be satisfactory (unpublished observations). The involvement of three or four clinicians at each examination also reduced the degree of imprecision in the organomegaly measurements, and previous analysis has demonstrated that interpretable changes in the measurements occurred within this cohort \[[@B13]\]. The major result from this analysis is that the outcome of treatment with praziquantel, combined with very limited S. mansonire-infection, depended strongly on where members of the cohort were resident. Previously, we demonstrated that children within the same cohort that had relatively high-level exposure to both S. mansoniand malaria presented with significantly larger spleens before treatment than children highly exposed to either parasite alone \[[@B14]\]. This contrasted with an observation on splenomegaly in school children exposed to malaria only in Ghana, where splenomegaly was observed to decline with distance from mosquito breeding sites \[[@B5]\]. Here, we report that children with relatively large spleens at baseline and living in the area with high levels of exposure to malaria still had the largest spleens 3 years after treatment, despite an overall reduction in their organomegaly. Hardness of enlarged spleens at baseline was most commonly observed amongst children living close to the river, irrespective of their level of pre-treatment intensity of S. mansoniinfection \[[@B14]\]. In the absence of follow-up data, the inference may have been that the hard consistency of enlarged spleens was primarily attributable to chronic malaria exposure. However, during the follow-up period, we observed a gradual and monotonic decrease in the prevalence of hard spleens that suggests the removal of S. mansoniinfection was the trigger for the improvement \[[@B13]\]. One possible explanation is that the hardening of the spleens was due to a synergistic effect of co-infection. Hyperplasia and congestion in the spleen associated with relatively high levels of exposure to malaria may have been sufficient to cause chronic enlargement of the spleen. Hardening and further enlargement may then have occurred as a secondary effect of congestion in the liver attributable to schistosome infection, leading to increased portal pressure and dilation of the splenic vasculature. Three years after treatment with praziquantel, and with very limited re-infection by S. mansoni\[[@B12]\], the prevalence of palpable spleens had diminished considerably. At all time points, the prevalence of hard spleens and the degree of splenomegaly were highest amongst children living in the sector with highest egg counts before intervention and highest anti-Pfs IgG3 responses. By removing S. mansoniinfection and limiting re-infection, we may have abrogated any synergistic effects of co-infection on the spleen, and thereby uncovered the residual burden of chronic exposure to malaria. Importantly, the rate of regression of MAL splenomegaly was not affected significantly by location, which suggests that the effects of praziquantel treatment on splenomegaly attributable to S. mansoniinfection are not dependent on the level of exposure to either malaria or S. mansoni. The effects of praziquantel treatment on hepatomegaly in this cohort were more subtle. There was gradual decline in the prevalence of hepatomegaly after praziquantel treatment \[[@B13]\] indicating that the removal of S. mansoniinfection was a critical factor for improvement. However, upon analysing hepatomegaly data across sectors, it emerged that children from the sector with relatively high anti-Pfs IgG3 responses, but relatively low S. mansoniegg counts, had the lowest rate of regression. A possible explanation of this observation is that an aetiological agent other than S. mansoniwas responsible for the observed hepatomegaly. A likely candidate is malaria infection. Enlargement of the liver in acute malaria infection is temporary and recedes rapidly after treatment \[[@B24]\]; however, studies of young children in Gambia have shown that repeated infection with malaria, perhaps when combined with other unidentified factors, can lead to the development of chronically enlarged livers \[[@B25],[@B26]\]. Our observations suggest that even if school-aged children are examined outside the malaria transmission season, they may still be affected by hepatomegaly attributable to malariainfection. Children from the sector with relatively low exposure to both infections experienced a significant decrease in liver enlargement along the mid sternal line when baseline data were compared with data from the third year follow-up. It is possible that hepatomegaly in these children was attributable to neither S. mansoninor malaria infection. Alternatively, because they were relatively lightly infected with S. mansoni, and experienced relatively low exposure to malaria infection, it is possible that they were more likely to regress in terms of hepatomegaly within the follow-up period. Conclusions =========== In conclusion, our observations lend further support to the hypothesis that severity of hepatosplenomegaly in Kenyan school-aged children is related to their degree of exposure to both S. mansoniand malaria infections. Specifically, although the degree of hepatomegaly or splenomegaly may be correlated with S. mansoniinfection, there is likely to be further exacerbation of the condition if a child is concurrently exposed to malaria. We have now observed that the apparent benefits of treating a case of hepatosplenomegaly with praziquantel may be reduced if a child lives in an area of relatively high exposure. We have therefore confirmed the long-standing, but untested, hypothesis that co-infections of S. mansoniand malaria may obscure the clinical evaluation associated with infection by either species. Our results also introduce the necessity of considering the level of exposure to malaria when evaluating the clinical outcome of praziquantel treatment for S. mansoniinfection. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MB conceived of, and conducted, the analysis, and drafted the manuscript. BJV and AEB performed clinical examinations and participated in the design of the study. CA and AO conducted clinical examinations. CHK, GK, JM, JHO and EM participated in the planning and execution of field activities. DWD participated in the design of the study and in fieldwork. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1741-7015/2/36/prepub> Acknowledgements ================ The authors gratefully acknowledge the co-operation of children and teachers from Mbeetwani, as well as the team of fieldworkers. Financial support was obtained from the Wellcome Trust and The Commission of the European Community\'s Science and Technology for Development Programme (INCO-DC contract IC18 CT97-0237 and INCO-DEV contract ICA4-CT-1999-10003). Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Map of the study area showing the course of the River Kambu, positions of cohort members households and the four sectors corresponding to areas of different exposure to S. mansoniand malaria infections. The accompanying table contains the median and 25--75%ile ranges for egg counts (e.p.g.) before treatment and anti-Pfs IgG3 OD responses for each sector. ::: ![](1741-7015-2-36-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Bar charts depicting temporal (by year) and spatial (by sector) variation in the prevalence of hard spleens (Figure 2a) and firm livers (Figure 2b). Estimates are from the baseline survey and surveys conducted up to three years after first treatment (Rx) with praziquantel. ::: ![](1741-7015-2-36-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Box and whisker plots depicting temporal (by year) and spatial (by sector) variation in spleen enlargement along the mid axillary line (Figure 3a) and the mid clavicular line (Figure 3b), as well as liver enlargement along the mid sternal line (Figure 3c). The horizontal line within each box represents the median; the lower and upper bounds of the box correspond to the 25- and 75%iles; the whiskers correspond to the range of non-outlying data. Measurements are from the baseline survey and surveys conducted up to three years after first treatment (Rx) with praziquantel. ::: ![](1741-7015-2-36-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Results of testing for variation in the prevalence of hard spleens and firm livers over time within each sector. Sectoris described in the legend to table 1. Q-- Cochrans Q statistic, N-- number of individuals with relevant data from all time points, p-- level of significance. ::: Hard spleens Firm Livers -------- -------------- ------------- ------- ------ ------- Sector N Q p Q p A 21 14.0 0.003 13.2 0.004 B 12 9.0 0.029 5.0 0.172 C 15 8.3 0.041 7.8 0.049 D 19 12 0.007 10.7 0.013 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Improvement in hepatosplenomegaly stratified by level of exposure to malaria and S. mansoni. Sectors are identified by letter and are described in the text. N~i~-- number of children in each sector with improved organomegaly at the 3 year follow up; N~t~-- total number of children examined at the third year follow-up; Z-Wilcoxon Rank analysis Z score indicating whether or not there was a decrease in the degree of organ enlargement between baseline and the 3 year follow up. A negative Z value and a corresponding pvalue of \<0.05 indicates a significant decrease in organ sizes within a given sector. ::: MAL splenomegaly MCL splenomegaly MSL hepatomegaly MCL hepatomegaly -------- -------------------- -------------------- -------------------- -------------------- ------- ------- ------------------- ------- ------- ------------------- ------- ------- Sector N~i~/N~t~ Z p N~i~/N~t~ Z p N~i~/N~t~ Z p N~i~/N~t~ Z p A 21/23 -3.81 \<0.001 15/23 -1.20 0.234 10/23 -1.54 0.125 7/23 -0.11 0.915 B 9/13 -2.58 0.010 5/13 -0.42 0.673 2/12 1.39 0.164 1/12 -1.54 0.123 C 9/15 -2.72 0.007 5/15 -1.13 0.260 8/15 -1.22 0.222 2/13 -128 0.201 D 18/20 -3.78 \<0.001 9/20 -2.72 0.007 12/20 -2.80 0.005 4/20 0.00 1.000 :::	PubMed Central	2024-06-05T03:55:48.403137	2004-9-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522803/", "journal": "BMC Med. 2004 Sep 27; 2:36", "authors": [ { "first": "Mark", "last": "Booth" }, { "first": "Birgitte J", "last": "Vennervald" }, { "first": "Anthony E", "last": "Butterworth" }, { "first": "Henry C", "last": "Kariuki" }, { "first": "Clifford", "last": "Amaganga" }, { "first": "Gachuhi", "last": "Kimani" }, { "first": "Joseph K", "last": "Mwatha" }, { "first": "Amos", "last": "Otedo" }, { "first": "John H", "last": "Ouma" }, { "first": "David W", "last": "Dunne" } ] }
PMC522804	Background ========== Lichen myxoedematosus (LM) represents a rare chronic skin disorder of unknown aetiology, which may often be accompanied by severe internal co-morbidity such as haematological involvement including paraproteinaemia, neurologic syndromes, gastrointestinal complications or cardiac abnormalities \[[@B1]-[@B4]\]. Clinically and histologically, LM is characterised by papular eruptions caused by an extensive dermal deposition of glycosaminoglycans \[[@B5]\]. Scleromyxoedema (SCL) is a variant of LM exhibiting erythematous, sclerotic and stiffed lesions beside lichenoid papules with only little tendency of spontaneous remission \[[@B6],[@B7]\]. Even though various experimental treatment modalities have been described, to date, curing of SCL is still not possible. The oncoming case presentation focuses on a progressive variant of SCL as referred to clinical, immunohistochemical and laboratory investigations, followed by low-dose methotrexate (MTX) and subsequent medium-dose ultraviolet A1 (UVA1) cold-light treatment. Case presentation ================= We report a 44-year-old Caucasian woman who initially presented in 2003 with a multitude of progressive lichenoid 2--4 mm papules starting two years ago, particularly marked on the forehead right above both eyebrows, on the dorsal aspects of the forearms, shoulders, legs as well as on the gluteal region, accompanied by severe pruritus (Fig. [1](#F1){ref-type="fig"}). Clinically, the papules were judged as discrete, circumscribed, erythematous to skin-coloured, firm skin eruptions associated with an induration and stiffening of the affected lesions. Furthermore, the patient complained about a progressive thickening of the glabella. Even though we cannot ensure continuous clinical deterioration, at the time of the initiation of therapy, there was no hint for a beginning stabilisation or even improvement of symptoms. Otherwise, she felt healthy and well. The general examination was without pathological findings. Neither a topical nor a systemic therapy was yet applied. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Lichenoid papules beside thickened skin on the dorsal aspect of the left shoulder. ::: ![](1471-5945-4-12-1) ::: Skin biopsies were taken from the left forearm/wrist, both legs and the left shoulder. Routine histological examination including haematoxylin-eosin, PAS labelling and Alcian blue staining revealed marked mucinous deposition within the upper and mid dermis beside an increased appearance of fibroblasts, collagen bundles and a discrete inflammatory infiltration. Additionally, immunohistochemical investigations were performed in order to enumerate CD4^+^(T helper cells), CD68^+^(macrophages), anti-mastcell-tryptase^+^(mastcells), decorin^+^(collagen fibril stability protein), MIB-1^+^(Ki-67^+^proliferating cells), CD20^+^(B lymphocytes), and FGF-R^+^(fibroblast growth factor receptor bearing cells) cells taking consecutive sections (Table [1](#T1){ref-type="table"}). A punch skin biopsy measuring 3 mm in diameter was taken from affected skin of the right forearm. 5 μm paraffin-embedded sections were cut, mounted on slides and routinely preserved. Prior to the single immunolabelling, different pretreatments were performed for antigen retrieval (Table [1](#T1){ref-type="table"}). The alkaline phosphatase anti-alkaline phosphatase (APAAP) technique using the labelled streptavidin-biotin (LSAB) method was used to enumerate immunopositive cells at an individual dilution (Table [1](#T1){ref-type="table"}) taking consecutive sections. The alkaline phosphatase fast red detection kit utilised a biotinylated secondary antibody that binds to the primary antibody. This step was followed by the addition of an streptavidin enzyme conjugate binding to the biotin present on the secondary antibody. Afterwards the specific antibody-secondary-antibody-streptavidin-enzyme-complex was detected using a precipitating enzyme reaction product. Each step was incubated for a precise time and temperature. The alkaline phosphatase was used as enzyme; the chromogene fast red could be visualised. Cells were evaluated semiquantitatively (absent (-), rare (o), moderate (+), frequent (++)) directly below the dermoepidermal junction. Immunopositive cells were evaluated \'blind\' separately in two view fields in a row (0.25 mm × 0.25 mm each) directly below the dermoepidermal junction resulting in a length of 0.25 mm and 0.50 mm in depth. In order to avoid a sampling error, a number of sections were randomly reevaluated by a second observer. In case of a significant difference, the sections had to be recounted by both observers. In brief, immunolabelling revealed occasional perivascular CD4^+^and CD20^+^lymphocytes located in the papillary dermis and a high number of anti-mastcell-tryptase^+^cells within the subepithelial perivascular infiltrate revealing a continuing decrease with increasing depth (Fig. [2](#F2){ref-type="fig"}). Simultaneously, an increased dermal expression of MIB-1^+^cells (Fig. [3](#F3){ref-type="fig"}), morphologically predominantly fibroblasts, within the upper and mid dermis and sporadic FGF-R^+^cells in an unspecific distribution could be detected. CD68 immunohistochemistry and intradermal decorin levels did not alter remarkably as compared to healthy controls (data not shown). Exact results of all immunohistochemical stainings are detailed in Table [2](#T2){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Overview about the performed immunohistochemistry (alkaline phosphatase anti-alkaline phosphatase (APAAP) technique using the labelled streptavidin-biotin (LSAB) method) ::: Antibody Source Pretreatment\* Dilution Incubation time ---------- -------------------------------------- ---------------- ---------- ----------------- CD4 Novocastra Loxo, Dossenheim, Germany H 1:60 30 min CD68 Dako, Hamburg, Germany P 1:25 30 min Tryptase Dako, Hamburg, Germany P 1:400 28 min MIB-1 Dako, Hamburg, Germany H 1:10 32 min FGF-R Oncogene Research, San Diego, USA N 1:10 30 min Decorin Oncogene Research, San Diego, USA N 1:10 30 min CD20 Novocastra Loxo, Dossenheim, Germany H 1:50 30 min \N = none, P = protease digestion, H = heat (microwave-3-step-technique) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Immunhistochemistry unveiling sporadic lymphocytes beside a high number of anti-mastcell-tryptase^+^cells within the subepithelial perivascular infiltrate. ::: ![](1471-5945-4-12-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Mucinous deposition of the upper and mid-dermis accompanied by an elevated occurrence of MIB-1^+^dermal fibroblasts. ::: ![](1471-5945-4-12-3) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Semiquantitative data\ of the immunohistochemical studies on a patient with scleromyxoedema ::: Antibody CD4 CD68 Tryptase MIB-1 FGF-R Decorin CD20 -------------- ----- ------ ---------- ------- ------- --------- ------ Upper dermis \+ o ++ \+ o \- o Mid-dermis \- \- \+ \+ \- \- \- Lower dermis o \- o o \- \- \- \*- = absent, o = rare, + = moderate, ++ = frequent ::: Complete laboratory measurements unveiled the following pathological results: leucocytes 10870 μL^-1^, lymphocytes 13.2%, IgG 2000 mg/dl. There was no increase in B cell count. Immunoelectrophoresis disclosed albumin 51.4%, alpha-2 globuline 10.5%, gamma globuline 22.9%. Cranial x-ray, x-ray of the thorax, ultrasound of the abdominal organs, electrocardiography and urinary investigations were unremarkable. Blood smear cytological evaluation revealed beginning qualitative but still no quantitative changes as defined by leukocytic aberrations pointing towards a leftward shift. Serum immunofixation demonstrated an IgG-kappa paraproteinaemia. No elevation of the IgG-lambda paraprotein was assessed. Bone marrow biopsies displaying reactive lymphoid infiltration including minimal extension of plasma cells with monoclonal immunoglobuline production provided evidence for monoclonal gammopathy of undetermined significance (MGUS) without distinct morphological characteristics of a plasmocytic plasmocytoma or plasmoblasts. Initially, 20 MHz ultrasound scanning producing cross section images of the skin was established in order to measure both structure and thickness of the skin at the dorsal aspects of the left wrist and the right forearm. The total thickness of the skin was measured from the entrance echo to the border between the dermis and the subcutaneous tissue. A cutaneous diameter of 2291 μm at the left wrist (lesional skin) and of 1106 μm at the right forearm (non-lesional skin) could be assessed. In our unit, an experimental treatment modality using oral MTX 12.5 mg once per week followed by a subsequent folic acid application on the following day for a 6-months-period was subsequently initiated. MTX was well tolerated by our patient. After the first three months, the continuous progress of skin lesions during the last two years could be stopped and our patient experienced subjectively an improvement and objectively a stable clinical outcome without new lesions. Subjective impression of amelioration could not be confirmed by means of ultrasound measurement. Within the following two months, no further improvement could be evaluated, whereas no further progression such as formation of new lesions or increase of stiffness could be observed. Due to the unsatisfying clinical results and declining acceptance by our patient, MTX treatment was stopped and a subsequent therapeutic attempt with medium-dose UVA1 cold-light phototherapy was initiated. Irradiation consisting of 50 J/cm^2^single-dose UVA1 (CL 300000 liquid, Photomed, Hamburg, Germany) was performed four times a week for three weeks followed by two times a week for further two weeks resulting in a cumulative dose of 800 J/cm^2^after five weeks. Meanwhile, the skin status again remained stable, whereas no improvement could be observed. Therefore, our patient broke up phototherapy. To date, skin condition has slightly worsened without any current treatment modality. Conclusions =========== The population prevalence of SCL is known to be extremely low. Skin lesions are characterised by an increased deposition of acid mucopolysaccharides within the papillary and upper reticular dermis \[[@B8]\]. Even today, aetiology and pathogenesis remain hypothetic. Aberrant dermal deposition of monoclonal paraproteins predominantly of the IgG subtype combined with elevated IgG serum levels indirectly stimulating fibroblast activity are frequently found in LM patients \[[@B9]-[@B11]\]. Nevertheless, fibromucinous lesions of LM without the presence of paraprotein accumulations have also been described \[[@B12]\]. Beside typical skin eruptions, LM might also be associated with severe internal and neurological abnormalities such as cardiac irregularities, paralysis, hemiparesis or even progress to coma \[[@B3],[@B4],[@B13]\]. Despite sporadic case reports introducing new therapeutic strategies in LM and SCL, common treatment modalities are still disappointing and unsatisfactory. Topical treatment including hyaluronidase and triamcinolone as well as systemic efforts by the use of corticosteroids, cyclophosphamide, electron-beam therapy, hydroxychloroquine, PUVA, extracorporeal photopheresis, plasmapheresis or high-dose intravenous immunoglobulin partly displayed only limited success in individual patients \[[@B12],[@B14]-[@B20]\]. In our patient, Alcian blue staining disclosed a remarkable deposition of mucinous material within the upper dermal layers combined with an increased appearance of proliferating MIB-1^+^and occasional FGF-R^+^fibroblasts in immunohistochemistry. Ki-67 (MIB-1) and fibroblast growth factors are involved in a variety of mitogen and proliferative activities \[[@B21]\]. Thus, the enhanced appearance of positive cells might represent an increased overall activation probably resulting in an aberrant release of mucopolysaccharides. Decorin contributes to the collagen fibril stability and high levels of decorin seem to be closely linked to dermal fibrotic stages as known from systemic sclerosis \[[@B22]\]. Here, an increased intradermal decorin expression could not be demonstrated. Simultaneously, in our patient the mucinous deposition was accompanied by a decreased presence of collagenous bundles. Interestingly, immunohistochemistry also revealed a number of CD4^+^and CD20^+^dermal inflammatory lymphocytes as well as anti-human mast cell tryptase^+^cells, which may profoundly contribute to mucinosis formation \[[@B23]\]. Unfortunately, we were not able to provide a longitudinal analysis of the different stainings due to missing consent of the patient to perform additional experimental biopsies within the course of therapy. By considering clinical appearance, laboratory findings, immunofixation, bone marrow biopsy and histological evaluation, our patient presumably exhibited a complex variant of SCL. Fibroblast activity was supposed to be increased reflected by a corresponding high number of MIB-1^+^and FGF-R^+^cells within the upper dermis beside a massive mucinous deposition. MTX therapy is approved in the treatment of malignant lymphoma. Additionally, low-dose MTX therapy has been established as a potent regimen in the treatment of T cell related skin diseases associated with a subsequent elevated fibroblast activation status (e.g. progressive systemic sclerosis) \[[@B24]\]. As in our patient paraproteinaemia of the IgG-kappa class, morphological signs of MGUS, proliferating dermal fibroblasts and a discrete T cell weighted lymphocytic dermal infiltration could be verified, we first decided to start a therapeutic attempt by the use of 12.5 mg MTX weekly in order to suppress local cellular activity following the promising reports about application of the anti-metabolite melphalan and the alkylating agent cyclophosphamide in previous studies \[[@B12],[@B14]\]. Follow-up examinations were performed monthly. Three months after initiation of MTX therapy, an encouraging stable clinical outcome as well as an decline of pruritus without further progression of the disease was observed. However, the 5-months-follow-up revealed no apparent improvement of skin status leading to the joint decision of breaking up MTX treatment. UVA1 phototherapy has been shown to be effective in a number of inflammatory and fibrotic skin disorders by the induction of T cell apoptosis, collagenase activity and antiproliferative pathways \[[@B25]\]. Therefore, we decided to initiate a second attempt by the usage of a common medium-dose UVA1 irradiation protocol. Nevertheless, our patient broke up this regimen even after a 5-weeks-period due to the absence of immediate clinical improvement and an unfavourable time/benefit ratio, while follow-up examinations during this time revealed a stop of the progress anyway. However, in order to interrupt the clinical progression or even therapy-resistance as reflected by our case presentation and by considering the stable skin conditions following UVA1 phototherapy, we are currently discussing a new therapeutic attempt applying extracorporeal photochemotherapy, as proposed by Krasagakis et al. \[[@B16]\], in order to stabilise or even improve the present slight aggravation without any potent therapy. List of abbreviations ===================== LM: Lichen myxoedematosus; SCL: Scleromyxoedema; MTX: methotrexate; UVA1: ultraviolet A1; MGUS: monoclonal gammopathy of undetermined significance Competing interests =================== None declared. Authors\' contributions ======================= F.B. participated in the design of the study, carried out the immunohistochemistry, performed the statistical analysis and drafted the manuscript. S.R. carried out 20 MHz ultrasound scanning. A.K. conceived of the study. M.F., M.S. and P.A. participated in histology, 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/1471-5945/4/12/prepub>	PubMed Central	2024-06-05T03:55:48.405660	2004-9-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522804/", "journal": "BMC Dermatol. 2004 Sep 22; 4:12", "authors": [ { "first": "Frank", "last": "Breuckmann" }, { "first": "Marcus", "last": "Freitag" }, { "first": "Sebastian", "last": "Rotterdam" }, { "first": "Markus", "last": "Stuecker" }, { "first": "Peter", "last": "Altmeyer" }, { "first": "Alexander", "last": "Kreuter" } ] }
PMC522805	Background ========== A consistent feature of aged and photoaged skin is the flattening of the epidermal-dermal junction, evidenced in histological sections as a loss of rete ridges and the disappearance of papillary projections. The depth of interdigitation of the retepegs and the dermal papillae decreases with age \[[@B1]\]. Using confocal scanning microscopy sections horizontal to the surface can be obtained optically in contrast to conventional transverse sections which are perpendicular to the surface. Comparison of confocal images with corresponding histological sections have been made by the group of Gonzales to validate the method \[[@B2]-[@B5]\]. In confocal images the dermal papillae appear as dark circles surrounded by bright reflecting rings of basal cells containing highly reflective pigment \[[@B6]\]. Each papilla contains a single nutritive capillary loop \[[@B7]\]. The density of these functional entities of dermal papillae containing a single nutritive capillary loop can be evaluated more accurately and precisely by confocal microscopy than by conventional histology. This is partly due to the avoidance of shrinkage artefacts after fixation for histological sectioning. Confocal microscopy is more suitable to construct a three dimensional image than conventional serial sections are \[[@B8]\]. In extremely aged skin, papillae virtually disappear and the junction with the atrophic epidermis is a straight line versus undulations in younger skin. The papillary dermis also thins along with a loss of capillaries. Grove described that the corneocytes in aged skin become larger as a result of decreased epidermal turn-over \[[@B9]\]. Similarly, cells of the granular layer become larger, indicating a slow epidermal turn-over in aged skin \[[@B6]\]. Unlike for example in mice, in humans like in primates in general and in guinea pigs vitamin C must be supplied exogenously in the diet. Diets deficient in vitamin C cause the multiple clinical signs of scurvy. Stones evolutionary treatise showed that a genetic mutation resulted in the loss of the ability of humans and some other animal species to synthesize vitamin C \[[@B10]\]. The main mechanism for the symptoms of scurvy seems to be the instability of non-hydroxylated forms of collagen, as vitamin C is needed for the hydroxylation of prolin \[[@B11]\]. There is evidence that topical vitamin C might be beneficial in several unrelated conditions. Topical vitamin C has been reported to improve wound-healing \[[@B12]\]. Roshchupkin has shown that topical vitamin C is protective against immediate effects of ultra-violet radiation on human skin leading to an increase in the dose required to induce erythema \[[@B13]\]. Topical vitamin C protects against ultra-violet-induced carcinogenesis \[[@B14]\]. The level of vitamin C in the skin decreases with age, especially in the epidermis \[[@B15],[@B16]\]. Topical vitamin C increases the mRNA levels of collagens I and III, and their processing enzymes in humans \[[@B17],[@B18]\]. Humbert showed the potential of ascorbic acid to improve the clinical appearance of photoaged skin and to reduce facial wrinkles \[[@B19],[@B20]\]. Retinoids, too, are known to improve the clinical appearance of photoaged skin and to promote the downgrowth of rete ridges, restoring the undulating dermo-epidermal interface \[[@B21]\]. A functional dermal-epidermal junction provides a better protection against mechanical stress which might detach the epidermis and lead to erosions \[[@B22]\]. The increased availability of oxygen and nutrients should also enhance repair after wounding and reverse the structural changes associated with photoaging. The height of the epidermal-dermal junction like it is measured as a parameter in conventional histology, can be determined by confocal microscopy only in a very inaccurate manner because of the horizontal orientation of the images. The aim of this study is first to establish a CLSM measurable parameter comparable to the height of the epidermal-dermal interface. It should be sensitive to find age-associated changes of the dermal-epidermal interdigitation. Second to examine the effects of a topical vitamin C containing moisturizing cream on the developed CLSM parameters. We used a dose of 3% vitamin C as it shows a good stability and a good release of the active ingredient in a special emulsion system \[[@B23]\]. Methods ======= In the first experiment (I), three groups of volunteers of different ages were compared measuring the density of the functional entities consisting of a dermal papilla and the nutritive capillary (papillary index). In a second experiment (II) a topically applied cream containing 3% vitamin C and its excipient were tested on the volar forearms of 33 volunteers. The effect of vitamin C cream was compared to excipient and to untreated control sites, respectively. (III) This experiment was repeated with a slightly different vehicle and for a two month treatment period only. Instrument ---------- The Vivascope 1000 (Lucid Inc., Rochester, N.Y.) is a commercially available confocal laser scanning microscope, which allows to examine human skin in vivo non-invasively. The system uses a Laser source with a wavelength of 830 nm, an illumination power up to 20 mW on the object and water immersion. In all three studies two different fields of view were used to obtain the papillary index (640 μm × 480 μm), and the projection areas of cells in the granular layer (230 μm × 88 μm). The lateral resolution of the instrument is 0.4 μm, the vertical resolution is about 1.9 μm. Parameters ---------- In confocal images of the forearm the epidermal-dermal interface shows dark round areas, the dermal papillae, surrounded by bright circles of basal cells, containing melanin granules and therefore reflecting strongly. Capillary loops are located in the centre of dermal papillae as black holes, often showing bright erythrocytes flowing through the capillary. The density of dermal papillae was evaluated by counting the papillae containing a capillary loop. 20 fields of view were investigated on each test site. The size of cells in the upper granular layer (A~gran~) was evaluated by saving images of the most apical plane of the epidermis that still showed living dark nucleated cells. Although in the same image you can have virtual sections of different cell layers, the cells belonging to this layer right underneath the stratum corneum can be identified by the characteristic morphologic features the cells display. The size of cells was analysed using the image analysis program image tool. At least 20 cells in at least three different images were analysed by trained examiners. Subjects and treatment ---------------------- \(I) In the first study three groups of female volunteers were studied: twelve, ages 18 to 25, eleven, ages 40 to 50 and twelve, ages 65 to 80 years. Informed consent was obtained. Images were obtained on the volar side of the right forearm during the winter months. \(II) In a second experiment a cream containing 3% vitamin C was compared to its vehicle on a group of initially 36 volunteers. 33 volunteers finally participated using it twice daily. Applications for four months to mid-volar forearms of healthy female postmenopausal subjects, ages 45 to 67 years (mean 55.3 years), phototype II and III. One area was left untreated. The double-blind study was conducted in winter. Informed consent was obtained. Evaluations were obtained at baseline and each month for four months, with one and two months follow-ups after stopping the treatment. \(III) To determine reproducibility a second study was conducted over an eight weeks treatment period with a slightly different vehicle. The parameter of interest was the papillary index. Informed consent was obtained. Measurements were made after 20 minutes of acclimatisation at 21°C and 50% relative humidity. Statistics ---------- Morphometric results are expressed in box-whisker-plots as the median +/- the quartile. Statistical analysis used the Student t-test for normally distributed data or the U-Test (Mann-Whitney) for the age comparison study. The Wilcoxon test was used for analysing the effect of the vitamin C studies. Statistical significance was established for the 95% confidence level (p \< 0.05). Results ======= \(I) The papillary index in the young group, with a median of 54.3 papillae per mm^2^, was significantly the highest of the three age groups. The middle aged group showed a significantly higher index, median 29.6 papillae per mm^2^, than the oldest group, median 24.7 papillae per mm^2^. The decrease of the papillary index was more pronounced between the younger group (18 to 25 years) and the middle-aged group (40 to 50 years) than between this group and the old group (65 to 80 years) (fig. [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Box-plot of papillae per mm^2^in three age groups ::: ![](1471-5945-4-13-1) ::: In the elderly group there were often large segments of flat epidermal-dermal junctions with a complete loss of papillae. In these regions the microvasculature consisted of horizontally orientated vessels having noticeably larger diameters than in young skin, probably venules rather than capillaries. A few curled remnants were occasionally found just beneath the atrophic epidermis. The vessel walls could not be resolved into endothelium, pericytes or smooth muscle cells by confocal examination, owing to low reflectiveness. \(II) From four weeks onwards ascorbic acid treated areas showed a significantly higher density of capillary containing papillae than untreated and vehicle treated sites (fig. [2](#F2){ref-type="fig"}, p-values in table [1](#T1){ref-type="table"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Box-whisker-plots of papillae per mm^2^during topical vitamin C; t0 = baseline, t1 to t4 = after 1 to 4 months, t5 and t6 = 1 and 2 months after stopping ::: ![](1471-5945-4-13-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### statistical analysis of the verum treatment study: p-levels, Wilcoxon Matched Pairs Test ::: vehicle vs verum untreated vs verum untreated vs vehicle ---- ------------------ -------------------- ---------------------- t0 58,70% 21,10% 36,40% t1 1,40% 0,35% 95,00% t2 0,03% 0,02% 62,60% t3 0,02% 0,03% 97,00% t4 0,01% 0,01% 87,30% t5 0,60% 15,30% 28,10% t6 4,28% 12,50% 77,30% ::: Statistically significant differences were also observed in the vitamin C treated sites from four weeks onward in comparison to the beginning of the study. The untreated as well as the vehicle treated site showed a slight, but significant increase (from median 25.3 to 26.9 papillae per mm^2^for the untreated and from median 22.7 to 30.3 papillae per mm^2^for the vehicle treated sites) over the treatment period as well. The significance of the difference between the verum and the two control sites were no longer evident two months after stopping treatment. Despite of the different median values for the areas at t0, no significant differences between these baseline values could be detected. It is noteworthy that the newly formed papillae could not be distinguished from the preexisting ones. Neither abnormal formations of blood vessels like multiple capillaries in a papilla, parallelisation of the vessels or highly enlarged diameters of the vessels nor signs of inflammatory reactions e.g. rolling or adhesion of leukocytes to the vessel walls were observed. The projection areas of the cells in the upper granular layer showed a significant decrease of about 16% from 472 μm^2^to 383 μm^2^in the vitamin C treated site, in comparison to baseline and the two control sites. Significant difference between vehicle and untreated controls compared with each other and with the initial status at the beginning of the study were not detected (fig. [3](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Box-whisker-plots of areas (μm^2^) of cells in the apical granular layer; t0 = baseline, t2 = after 2 months of treatment ::: ![](1471-5945-4-13-3) ::: \(III) The increase of the density of the papillae under influence of vitamin C was reproduced by repetition of the treatment schedules, ruling out the likelihood of changes due to chance. Discussion ========== The decrease of the height of the epidermal-dermal junction is a well described histological finding in aged and photoaged skin \[[@B22],[@B24]\]. We conclude from our data that ageing results in a damage of the papillary dermis and of the vascular network in this layer and a decrease of the papillary density over time. Vitamin C has been reported to improve the clinical appearance of photoaged skin and to enhance the synthesis of composite elastin fibres and of collagen \[[@B19]\]. Our data indicate that the topical application of vitamin C partially restores the anatomical structure of the epidermal-dermal junction in young skin and increases the number of nutritive capillary loops in the papillary dermis close to the epidermal tissue in the aged skin of postmenopausal women. The increase in the density of papillae after vitamin C treatment can not be interpreted only as a sign of epidermal hyperplasia with an enlarged area covered with basal layer and the epidermis growing down into the dermis. More likely it is linked to a restructuring of the papillary dermis, as the top of the newly formed papillae and the capillaries are localized above the average height of the plane basal layer seen predominantly in aged skin. This suggests restoration to a more normal functional state of the epidermal-dermal interdigitation and of the overlying epidermis. In aged skin, the cells in the apical granular layer show larger projection areas than in the skin of younger individuals \[[@B8]\] similar to the enlargement of the size of corneocytes with age described earlier by Grove and Kligman \[[@B9]\]. Kligman shows an inverse relationship between turnover and the size of corneocytes \[[@B9]\]. A similar relationship could be shown for the size of cells in the granular layer \[[@B25]\]. Parallel to this, the smaller size of granular layer cells in the vitamin C treated areas can be interpreted as a sign for higher proliferative activity of the epidermis. Another implication of the increase in papillae is that new blood vessels are formed during the treatment with vitamin C. The newly formed blood vessels show a normal anatomical structure in confocal microscopical examination and are apparently integrated in a healthy vascular architecture. In the confocal images there are no signs of pathologic changes of the vasculature like, for example, increased diameter, parallel orientation or clew of vessels in a papillae. No perivascular infiltrations of lymphocytes could be observed during the treatment period. The mechanism, by which Vitamin C restores dermal papillae is unknown. Conclusions =========== These are early results which strongly suggest that topical vitamin C may have important anti-aging effects in correcting the structural and functional losses associated with skin aging. Competing intrests ================== The studies were funded by the Beiersdorf AG, Germany. Parts of the results presented in the manuscript are subject of patents pending by Beiersdorf AG. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-5945/4/13/prepub> Acknowledgements ================ The authors want to thank Sven Clemann, Michaela Priebe, Patricia Sauermann-Wehde and Manuela Fuerstenau for the technical support of the studies. Also thanks to Volker Schreiner, Alexander Filbry, Ralph Schimpf, Roger Wepf, Stephan Teichmann and Joachim Ennen for encouraging us to pursue the idea. We want to especially thank Prof. Dr. A Kligman for the encouraging review of our work.	PubMed Central	2024-06-05T03:55:48.407272	2004-9-29	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522805/", "journal": "BMC Dermatol. 2004 Sep 29; 4:13", "authors": [ { "first": "Kirsten", "last": "Sauermann" }, { "first": "Sören", "last": "Jaspers" }, { "first": "Urte", "last": "Koop" }, { "first": "Horst", "last": "Wenck" } ] }
PMC522806	Background ========== With the successful completion of the Human Genome Project (HGP), we are entering the post genomic era. Facing mass amounts of data, traditional biological experiments and data analysis techniques encounter great challenges. In this situation, cDNA microarrays and high-density oligonucleotide chips are novel biotechnologies as global (genome-wide or system-wide) experimental approaches that are effectively used in systematical analysis of large-scale genome data. In recent years, with its ability to measure simultaneously the activities and interactions of thousands of genes, microarray promises new insights into the mechanisms of living systems and is attracting more and more interest for solving scientific problems and in industrial applications. Meanwhile, further biological and medical research also promoted the development and application of microarray. Typical issues addressed by microarray experiments include two main aspects: finding co-regulated genes for classification based on different cell-type \[[@B1]\], stage-specific \[[@B2],[@B3]\], disease-related \[[@B4]-[@B6]\], or treatment-related \[[@B6]-[@B8]\] patterns of gene expression and understanding gene regulatory networks by analyzing functional roles of genes in cellular processes \[[@B9],[@B10]\]. Here we focus on the former, especially on tumor classification using gene expression data, which is a hot topic in recent years and has received general attention by many biological and medical researchers \[[@B11]-[@B19]\]. A reliable and precise classification of tumors based on gene expression data may lead to a more complete understanding of molecular variations among tumors, and hence, to better diagnosis and treatment strategies. Microarray experiments usually generate large datasets with expression values for thousands of genes (2000\~20 000) but not more than a few dozens samples (20\~80). Thus, very accurate classification of tissue samples in such high-dimensional problems is difficult, but often crucial, for successful diagnosis and treatment. Several comprehensively comparative and improved methods have been proposed recently \[[@B20]-[@B22]\]. In this paper, we introduce a combinational feature selection method using ensemble neural networks to remarkably improve the accuracy and robustness of sample classification. In recent years, several researchers have used ensemble neural networks for tumor classification based on gene expression data \[[@B12],[@B23]\]. Khan et al. \[[@B12]\] used neural networks to classify 4 subcategories of small round blue-cell tumors. By using 3750 networks generated by three fold cross-validation 1250 times and using the list of 96 most influential genes as the inputs, they reported very excellent results based on their dataset. Also O\'Neill and Song \[[@B23]\] used neural networks to analyze lymphoma microarray data and can predict the long-term survival of individual patients with 100% accuracy based on the datasets published by Alizadeh et al \[[@B18]\]. Both of them are very good work in microarray data analysis using neural networks. In this paper our motivation lies in that by combining various feature selection mechanisms we can avail of more information of samples for classification and by using ensemble neural networks we can more effectively combine these features and improve the stability and robustness of answers. So the most important distinctions between our work and these above two citations are that by using combinational feature selection we can penetrate various different profiles of the samples and can avail of more information for classification, and also these neural networks can work in a parallel way unlike those two papers. In the same time, unlike their work based on some certain dataset, we can get improved, at least comparable results on a wide range of different datasets. In the following section, we provide detailed illustration and comparison of our new method. Results ======= The general framework and implementation of our method ------------------------------------------------------ The flowchart of our method can be seen in Figure [1](#F1){ref-type="fig"}. When we obtain the microarray raw data based on a certain classification problem, first we need to preprocess them in order to be beneficial for further analysis. Broadly defined, pre-processing includes the planning and design of experiments, the acquisition and preprocessing of images, data transformation, data inspection, and data filtering. In this paper we avail of the publicly available datasets in <http://sdmc.lit.org.sg/GEDatasets/Datasets>, so we simplify this step and only use all datasets exactly as we found them in their transformed data. Due to the characteristic of small sample numbers in microarray data, in order to improve the accuracy, robustness and generalization of issue classification, we apply bootstrap mechanism to resample 100 iterations. During each iteration, we input the resample training data into three cooperative and competitive neural networks, and then by averaging their decisions, the neural network set can output their discrimination. From Figure [2](#F2){ref-type="fig"}, we can clearly understand the architecture of these three neural networks. After obtaining the transformed resampling data, we extract and select features respectively based on ranksum test, PCA, clustering and t test. Ranksum test (also named Wilcoxon/Mann-Whitney test) is a nonparametric test, which does not take values into account and only calculates their scores purely based on rank information. We chose the top-ranked 30 genes identified as differentially expressed between the two types of tissues according to the ranksum test with the highest confidence (here using training data) as the first network input. At the same time, we used PCA to extract the principle components of all genes and used the top 15 principle components as the features to input another neural network. Also, we used Jaeger\'s \"Masked out Clustering\" ideas to group all the genes into 50 clusters and then used a t test to obtain the top 30 significant genes. Here we assume that each cluster can belong to the same pathway, genes which are co-expressed or are coming from the same chromosome. In this way, we can prefilter the gene set and drop genes that are very similar or highly correlated; that is, we can select the more significant genes for our discrimination as the third network input. More information about feature selection can be found in the methods section later. Based on these above three kinds of features we selected as the input, we construct and train three neural networks. Here we adopt simple one-hidden-layer feed-forward networks, which have 10 hidden units and one output unit for binary classification problem. As for multi-class problems, we can accordingly change the number of output units. Because each of these three networks adopts different feature selection mechanism as inputs, these inputs respectively reflect different aspects of samples, that is, different feature space in discriminative problems. We believe that this strategy of feature selection for issue classification reflects more profiles of different classes and will be able to obtain more accurate solution. Actually each of three networks is just like an expert holding a different judgment mechanism. Through averaging the confidences of three experts\' answers, we can get the answer of this expert system. In this way, we not only can get the confidence of each expert, also we can judge the weight of each type of features in the answer. Finally, through competitive neural networks the robustness of this problem will be improved greatly. After completing the 100 iterations, we can get 100 individual answers about the problem. In this situation, how to combine these answers into one more precise result is still a problem. Here, we simply use majority voting to combine the result and then give the ultimate solution about this classification problem. As noted above, here we adopt the soft-voting mechanism, that is, we can combine the confidence of each net. All the implementations of our framework were written in Matlab, using the hardware platform of a PC running 2.4 GHz. Datasets illustration --------------------- In this section, simple illustrations of the datasets we used in this paper for exploring the performance of our classification are given. The datasets in our paper have been downloaded from the following website: <http://sdmc.lit.org.sg/GEDatasets/Datasets>. We adopted their transformed data format for further research. All datasets we used can be reduced to three categories: binary class with testing samples, binary class without testing samples and multiple class problem. Here we classify samples into binary class with testing samples and without testing samples just according to the reference authors for each dataset. One important reason is that in this way we can easily compare our result with others based on the same training and testing sets. These datasets are shown in Table [1](#T1){ref-type="table"}. We use the three datasets below as the example of the first category, for which performance of our classification can be tested using the error ratio of testing samples. ### ALL-AML leukemia The training dataset consists of 38 bone marrow samples (27 ALL and 11 AML), with 7129 probes from 6817 human genes. Also, 34 samples testing data is provided, with 20 ALL and 14 AML. More information and raw data can be found in Golub et al. \[[@B11]\]. ### Lung cancer The dataset can be reduced to the problem of classification between malignant pleural mesothelioma (MPM) and adenocarcinoma (ADCA) of the lung. The training set contains 32 tissue samples, which consists of 16 MPM and 16 ADCA and the testing samples are constitutive of 15 MPM and 134 ADCA. Each sample is described by 12533 genes. More information about this dataset can be found in Gordon et al. \[[@B17]\]. ### Prostate cancer For the prostate cancer dataset, detailed explanation and raw data is available in Singh et al. \[[@B5]\]. This dataset consists of 102 training vs. 34 testing (Tumor versus Normal classification) samples. The training set contains 52 prostate tumor samples and 50 normal samples with around 12600 genes and the independent test sets consist of 25 tumor and 9 normal samples. Another three recently popular datasets have been used as the representative of the second category. Using these kinds of datasets, we apply cross-validation to validate our classification performance. ### Types of diffuse large B-cell lymphoma This dataset is used for discriminating distinct types of diffuse large B-cell lymphoma (DLBCL) using gene expression data. There are 47 samples, 24 of them are from \"germinal canter B-like\" group while the rest 23 are form \"activated B-like\" group and each sample can be described by 4026 genes. More detailed explanation can be found in Alizadeh et al. \[[@B18]\]. ### Ovarian cancer The goal of this significant experiment is to identify proteomic patterns in serum that distinguish ovarian cancer from non-cancer. The proteomic spectra were generated by mass spectroscopy and the dataset provided here is 6-19-02, which includes 91 controls (Normal) and 162 ovarian cancers with 15154 molecular mass / charge (M/Z) identities. Here we use the transformed normalization data in <http://sdmc.lit.org.sg/GEDatasets/Datasets>. More information can be found in Petricoin et al. \[[@B6]\]. ### Colon tumor The dataset Contains 62 samples collected from colon-cancer patients. Among them, 40 tumor biopsies are from tumors (labelled as \"negative\") and 22 normal (labelled as \"positive\") biopsies are from healthy parts of the colons of the same patients. Two thousand out of around 6500 genes were selected based on the confidence in the measured expression levels. Raw data and more information can be found in Alon et al. \[[@B14]\]. Finally, we can generalize our method from binary class to multi-class problems. In this paper, we evaluate the performance using the dataset below. ### MLL\_leukemia This dataset contains training data consisting of 57 leukemia samples (20 ALL, 17 MLL and 20 AML) and testing data consisting of 4 ALL, 3 MLL and 8 AML samples. We adopted the transformed data from <http://sdmc.lit.org.sg/GEDatasets/Datasets>. More information can be seen in Armstrong et al. \[[@B15]\]. Our results ----------- First we primarily focus on the binary class problem. Because most of problems can be reduced to binary class problems, such as diseased vs. normal, survival vs. lethal, two opposite subtypes of some diseases and so on. Finally we generalize our classifier to multi-class application. In this paper, we evaluate the performance of different classification methods using predictive accuracy, which can be defined as: ![](1471-2105-5-136-i1.gif) Here, TN1,TN2,...,TNnrespectively denote the correct classification numbers of the samples belonging to a corresponding class; totalnumrepresents total sample numbers. ### The results of binary classification with testing samples For the first category of the datasets, we evaluate the performance of our classifiers using predictive accuracy of testing samples compared with the best performance of the current available methods. In this paper we use bagging to resample just as Tan and Gilbert \[[@B24]\], and we also compared our results to those using their bagged decision trees. In Table [2](#T2){ref-type="table"}, we described the recognition rate of our methods compared with the best classifiers of our knowledge for each certain dataset and bagged decision trees proposed by Tan and Gilbert \[[@B24]\]. From this table, it is clear that our results are remarkably better than others based on these several datasets. For the most popularly used AML-ALL leukemia dataset, to our knowledge, the best classifiers of this dataset can be found in \[[@B25]-[@B27]\], which can predict the results with 97.1% accuracy. However, we designed the classifiers using our methods based on 38 training samples, 0 error number of 34 testing samples can be obtained from our classifier. In the same way, we trained our ensemble of neural networks using 32 training sets of lung cancer and then predicted the 149 separate testing sets still with 0 error number. And three (1:2) testing error numbers can be reached using methods by Li et al. \[[@B28]\], which is the best performance corresponding to this dataset of our knowledge. For the third prostate cancer dataset, after training the classifier using 102 training sets, only one wrong classification can happen using our ensemble neural networks to predict the 34 separate testing samples. We did not find a more accurate classification result except for the bagged decision trees in \[[@B24]\] based on this dataset, so here we think that is the best result. In this sense, a great improvement in predictive accuracy can be obtained by using our method. In order to further validate the effectiveness of predictive accuracy, we also performed the leave-one-out cross-validation (LOOCV) respectively only on the above three training samples. We also obtained the 100% accuracy both on the AML-ALL leukemia dataset and the lung cancer dataset, which are the same results as using individual testing samples. At the same time, 96.08% accuracy can be got based on the prostate cancer dataset, which is a little lower than using individual testing samples. For the purpose of comparison, we also list these results in Table [2](#T2){ref-type="table"}. Thus, we conclude that our performance evaluation is credible. From the results of the above three testing datasets, we can also see that many different classifiers obtain the best results when they concern some certain dataset, but there is still no general best strategy for tumor classification problems based on a wide range of different datasets. Furthermore, from Figure [3](#F3){ref-type="fig"}, it is clear that our method is superior to the traditional bagging decision trees. Thus, we conclude that by using our method a more general accuracy improvement can be achieved for tumor classification. ### The results of binary classification without testing samples Without separate testing samples, we cannot evaluate the performance of our classifiers with the predictive accuracy of testing samples in the same way as above. Many performance evaluation methods have been proposed, of which various cross validations are most popularly used, such as 3-fold cross validation, 10-fold cross validation, leave-one-out cross-validation (LOOCV), and others. Here, we used the leave-one-out cross validation (LOOCV) to evaluate the performance of ours based on these available datasets. For further comparison with recent published methods based on the same datasets, we also perform 10-fold cross validation just as they used in their research. In Table [3](#T3){ref-type="table"}, we list the predictive accuracy of our methods using 10-fold cross validation and LOOCV respectively and the corresponding results of other methods based on the same dataset and the same evaluation mechanism. These comparisons based on data in Table [3](#T3){ref-type="table"} are shown in Figure [4](#F4){ref-type="fig"}. In the first data column of Table [3](#T3){ref-type="table"}, we show our predictive accuracy 97.87% and 95.74% by LOOCV and 10-fold cross validation respectively. But unfortunately, we did not find the corresponding result based on this dataset. Cho et al. \[[@B25]\] artificially divide the dataset into 22 training samples and 25 test samples, and their best classification result is 96%. For the purpose of comparison, we also use the same strategy as Cho et al.\'s \[[@B25]\] and in Figure [4](#F4){ref-type="fig"} we can see that 98% predictive accuracy obtained by our method is a little better than theirs. O\'Neill and Song \[[@B23]\] used neural network to get very good result based on the lymphoma dataset. But here the dataset we used is based on different subset and we can\'t compare our result with theirs. For the ovarian dataset we used, Liu et al. \[[@B21]\] reported that 100% predictive accuracy can be obtained running 10-fold cross validation on all 253 samples under an all-χ^2^feature selection heuristic and support vector machine (SVM). In our method, only 75 features in total were used, and 99.21% and 98.82% accuracy was obtained respectively by LOOCV and 10-fold cross validation. Thus, we think that our method is comparable to theirs to some extent. For the colon tumor dataset, we found that 85.48% predictive accuracy is the best classification result obtained in Dettling et al. \[[@B27]\], where they used various boosting algorithms and adopted leave-one-out cross validation (LOOCV). As shown on Figure [4](#F4){ref-type="fig"}, compared with our predictive accuracy by LOOCV, a significant accuracy improvement was obtained by using our method. Our result by using 10-fold cross validation also is shown in Table [3](#T3){ref-type="table"}. From the results of the above three datasets, we can see that our method is better, or at least comparable to current other best methods. Also, we need to note that these named best methods can get the best results based on certain datasets but may get worse results based on other datasets (Here we omitted the concrete comparisons on wide range datasets; correlated information can be found in our references); however, general performance improvement can be obtained using our method. ### The result of multi-class problem Finally, we generalize our method from binary class to multi-class problems. After minor adjustments to the corresponding parameters of our framework, we obtained 100% classification result of the above-mentioned multi-class dataset -- MLL\_Leukemia dataset, that is, 4 ALL, 3 MLL, 8 AML of 15 test data can be predicted correctly. Similarly, 100% accuracy also was obtained by Li et al. \[[@B28]\]. In this way, we conclude that our method also is fit to multi-class problems, and the classification result is comparable to other methods. Discussion ========== In this paper, we introduce a combinational feature selection and ensemble neural network method for the classification of gene expression data. On a wide range of recently published datasets, our method performs better, or is at least comparable to, the current best methods of our knowledge. As a further test, we randomly selected genes of the same amount as the feature instead of any of the three individual selected features in our research and then used the ensemble neural networks based on these features for classification again. The apparently worse discrimination power can be seen in this strategy. Moreover, we also used the output of a unitary network based on all the same features as the ultimate classification result and the result was also worse than ours. Thus, we believe such remarkable performance improvements of our method are due to the fact that our combinational feature selection mechanism induced more useful information for discrimination, and the ensemble neural network framework improved the stability, robustness and generalization of learning. We performed simple majority voting mechanism to combine the individual networks produced by bagging and got a more accurate solution. The advantage of the ensemble is to reduce the variance, or instability of the neural network, and avoid the error surface of neural network training being trapped into local minima. The ensemble model tends to cancel the noise part as it varies among the ensemble members, and it tends to retain the fitting to the regularities of the data. In this paper, our ensemble neural network model has 100 members; However, further research is needed to determine how many members working together can reach the best performance. In this paper, we focused on classification problem, so we didn\'t give a detailed analysis about how the importance of each different gene we select and the interaction between them influenced the diseases, which is a very important issue for application and will be researched in our future work. Note that the only drawback of our approach is the problem of increasing computational complexity and the fact that it consumes a little more time than others. However, considering the lost caused by wrong prognosis or diagnosis of disease, we believe that the remarkable improvement in corresponding accuracy deserve these costs. Conclusions =========== By aggregating various information and ensemble neural networks, we reached a more accurate classification decision based on several datasets. We think that making full use of all available information will more clearly elucidate the latent mechanisms of many diseases. For example, we can combine various imaging techniques, such as CT, MRI, PET and others, which can detect the change of phenotype for the corresponding disease, with microarray data for further research. In this way, we can recognize the nature of various life phenomena both from macro and micro viewpoints. Also, we can retrieve the information of genes that are used in microarray, such as gene functions and gene locations. In this way, we can make use of prior knowledge combined with the microarray data for further research. Methods ======= Feature selection ----------------- Feature selection is one of the most important issues in classification, which is a transformation process of observations in order to obtain the best pathway for getting to the optimal solution. At the same time, it can reduce the complexity of the data to make it more comprehensible. It is particularly relevant for microarray datasets with thousands of features because it has been reported that many diseases, especially tumors, have never been caused by a single gene mutation but are the result of a series of gene changes. Such genes are highly relevant to the studied phenomena of diseases. On the other hand, the expression levels of many other genes may be irrelevant to the distinction between tissue classes. We can say that the extraction and selection of features determine the ultimate performance of classifiers. Both for cost and for biological insight, making full use of the most informative genes and finding small feature sets with high classification accuracy are very essential. At the same time, highly informative genes that are part of known biochemical pathways give insights into the processes that underlie the differences between classes, and those of unknown function suggest new research directions. Some classifiers, such as trees, perform automatic feature selection and are relatively insensitive to the variable selection scheme, but most classifiers need to perform feature selection first. So far, various feature selection schemes have been used in microarray data analysis, such as the most popular method of selecting the top-ranked genes based on various different scores (Euclidean distance, correlation coefficient, mutual information, signal to noise ratio) \[[@B9],[@B11],[@B22],[@B25],[@B29]\]. These feature selection methods gain better results on certain datasets, and the selective informative genes are the marker genes providing more useful information for further diagnosis and treatment. However, a problem with the above approaches is that they tend to select more correlated features so as not to provide more useful information for the purpose of classification. Li et al. \[[@B28]\] conclude that sometimes low-ranked genes are found to be necessary for classifiers to achieve perfect accuracy. It is conceivable that these useful low-ranked genes might have some relations with some important biological pathways and might have a vital influence on some diseases. Just selecting top-ranked genes will inevitably lose essential information. In order to compensate for this shortcoming, Jaeger et al. \[[@B22]\] proposed an improved gene selection for classification of microarrays. They demonstrated that the traditionally selected genes based on top-ranked scores are usually highly correlated, and they solved that problem through retrieving groups of similar genes first and then applying test-statistic to finally select genes of interest from these groups. In this way, the selected genes can correspond with some biological insights and might give out more accurate prediction about disease. The difficulty of this method lies in determining how many clusters and how many genes might directly correspond to the pathway on certain problems. Also, many researchers get the first several principle components by using PCA or SVD as the selected features, which captures most variation between samples and to some extent can obtain better results \[[@B12],[@B30]-[@B33]\]. However, principle components cannot provide comprehensible rules to help elucidate the scheme of the related disease because it can be due to noise as well as true difference in expression and we do not know how many genes to pick. Just as we alleged above, in such a high-dimension space, finding accurate and significant features (genes) is very essential for classification, for cost savings and for biological insights. However, it has been verified that differently selected features reflect different aspects of the dataset and some selected features can obtain better solutions on certain problems. This is because one feature selection mechanism corresponds to one different artificial hypothesis, but which hypothesis is most near to the true hypothesis on a special problem is unknown to us. Here, we propose combining the above mentioned several feature selection methods to reflect different profiles of samples in order to obtain more useful information for classification and to produce a good approximation to true hypothesis by averaging the different hypotheses. In this paper, we select features using wilcxon\'s ranksum test \[[@B34]\] to get the top-ranked genes, use PCA \[[@B31]\] to obtain the principle components as the feature, and use Jaeger\'s clustering method to group the whole genes into different clusters, and then select the top-ranked genes by t-test \[[@B34]\] scores from these groups. After picking these features from gene expression data, how to make full use of these features for further accurate classification is still a problem. Detailed illustration of our strategy is given below. Ensemble neural networks ------------------------ An Artificial Neural Network (ANN) is an information-processing paradigm that is modeled on biological nervous system, which is composed of a large number of highly interconnected processing elements (neurons) working in unison to solve specific problems. In fact, since the basic model was proposed, various improved algorithms and theories have already been successfully applied in many fields. Because neural networks are best at identifying patterns or trends in a large amount of data with little theory, they are well suited for prediction or forecasting needs. That\'s just the case for microarray data. However, instability and little intrinsic knowledge of neural networks are obstacles to its further generalized application in some specific problems. Here we ensemble multiple networks in an attempt to solve this problem to some extent. Since multi-net systems were introduced by Sharkey in 1996 \[[@B35]\], the combination of a number of neural networks has been widely applied in many fields. Because combining the outputs of several neural networks into an aggregate output often gives improved accuracy over any individual output, the objective of this kind of ensemble module is to solve problems that are difficult for a single neural network, and is to combine the individual outputs to achieve better generalization. Remarkable advantages of the ensemble compared with a unitary network have been demonstrated previously \[[@B36],[@B37]\], one advantage of which is that it can to some extent ease the obstacles mentioned above and can improve the stability of neural network decisions. Generally speaking, in neural networks ensemble, two problems need to be resolved: how to generate the individual network and how to combine them together. Using bootstrap or boosting resample mechanism to obtain the individual network is the most popular method to solve the first problem. Bootstrap is the most popular resample mechanisam of sampling with replacement, therefore some observations are duplicated and some are omitted. Boosting means to boost a \"weak\" learning algorithm into a \"strong\" learning algorithm. Their differences are that using bootstrap can resample uniformly and can get the individual network immediately, but boosting weights every sample in each iteration and must generate the individuals in sequence. Several recent published papers claimed that adaboost-the basic boosting algorithm is not fit to microarray data analysis \[[@B20],[@B26]\], and some improved boosting have been made to increase the accuracy to some extent \[[@B26],[@B27]\]. With no exception, they perform boosting in conjunction with decision trees. Here, we perform our resample mechanism using bagging, that is, bootstrap aggregating, which has been shown to work well in the presence of noise \[[@B24]\]. Due to the noisy fact of microarray data, here we use bagging to resample. As a further validation, we also used adaboost instead of bagging to resample in order to construct the individual network and the result is worse than bagging networks. As to the second problem, many ensemble mechanisms have been researched in recent years; for example, improvements in performance can result from training the individual networks to be decorrelated with each other \[[@B38]\] with respect to their errors. In this paper, we only adopt the majority voting, the basic ensemble method, to obtain the ultimate output result. Note that here we use the soft-voting mechanism, that is, the confidence of each net output is applied as voting value, rather than unit or zero. Considering the complexity and noise of microarray data, it\'s very difficult to get a perfect solution using a unitary neural network based on some certain selected features. At the same time, accurately extracting and selecting the most informative genes is also very difficult using individual available methods. Thus, in this paper we attempt to combine multiple modes of information available from gene expression data using neural networks ensemble to get a better solution. We presented a cooperative and competitive neural network system that each of nets has the same architecture and topology, and each can respectively learn to classify a set of patterns based on partial information of the patterns and then by combining their classification results we can get a more precise result. The detailed framework of combing various features and neural networks ensemble and its implementation methods are discussed below. Authors\' contributions ======================= BL carried out the design of the method and performed the related analysis. QC participated in discussions of algorithms and manuscript preparation. TJ and SM instructed the whole study. All authors read and approved the final manuscript. Acknowledgements ================ We are very grateful for the public accession of gene expressed database <http://sdmc.lit.org.sg/GEDatasets/Datasets>, which are the materials of this research. We are also very thankful to Elizabeth Budy, an English teacher of Chinese Academy of Science, for her careful reading and editing the manuscript. We thank the reviewers a lot for many good advices in our revision process. This work was partially supported by the Hundred Talents Programs of the Chinese Academy of Sciences, and the Natural Science Foundation of China, Grant No. 60172056 and 60121302. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The whole flow chart ::: ![](1471-2105-5-136-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Three cooperative and competitive neural networks ::: ![](1471-2105-5-136-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Comparing predictive accuracy of 3 separate testing samples with other methods ::: ![](1471-2105-5-136-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Comparing predictive accuracy of 3 datasets without testing samples with other methods ::: ![](1471-2105-5-136-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Gene expression datasets used in this paper ::: Dataset Number of genes Training samples Testing samples References ------------------ ----------------- ------------------ ----------------- ------------------------ ALL-AML Leukemia 7129 38 (27:11) 34 (20:14) Golub et al (1999) Lung Cancer 12533 32 (16:16) 149 (15:134) Gordon et al (2002) Prostate Cancer 12600 102 (52:50) 34 (25:9) Singh et al (2002) DLBCL 4026 47 (24:23) 0 Alizadeh et al (2000) Ovarian Cancer 15154 253 (91:162) 0 Petricoin et al (2002) Colon Tumor 2000 62 (40:22) 0 Alon et al (1999) MLL\_Leukemia 12582 57 (20:17:20) 15 (4:3:8) Armstrong et al (2002) All these datasets are downloaded from <http://sdmc.lit.org.sg/GEDatasets/Datasets> ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### The predictive accuracy of testing samples ::: ALLAML Leukemia Lung cancer Prostate cancer --------------------------- --------------------- ----------------- --------------------- Bagged decision trees 91.18% 93.29% 73.53% The best methods 97.06% 97.99% 73.53% Our methods 100% 100% 97.06% LOOCV on training samples 100% 100% 96.08% \* Note that the row of the best methods refer to the different method in different datasets ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The predictive accuracy by LOOCV and 10-fold CV ::: DLBCL Ovarian cancer Colon tumor ------------ --------------- ------------- ---------------- ------------- LOOCV Other methods --- --- 85.48% Our method 97.87 % 99.21% 91.94% 10-fold CV Other methods --- 100% --- Our method 95.74% 98.82% 90.32% \* Note that the two rows of other methods refer to the different best method in different datasets :::	PubMed Central	2024-06-05T03:55:48.408882	2004-9-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522806/", "journal": "BMC Bioinformatics. 2004 Sep 27; 5:136", "authors": [ { "first": "Bing", "last": "Liu" }, { "first": "Qinghua", "last": "Cui" }, { "first": "Tianzi", "last": "Jiang" }, { "first": "Songde", "last": "Ma" } ] }
PMC522807	Background ========== One particular family of cuticular proteins constitutes one of the largest multigene families known in insects \[[@B1]\]. Unrelated cuticular proteins are also numerous within a single species \[[@B2],[@B3]\]. This diversity of cuticular proteins is extraordinary when one considers that chitin, the other principal constituent of cuticle, is a simple filamentous polymer of N-acetylglucosamine. Over 60 sequences have been obtained from proteins extracted from arthropod cuticles freed from adhering cells, primarily through the work of Svend Andersen and his colleagues in Copenhagen. An additional 9 have been extracted from cuticle and had their N-terminal sequences determined \[[@B2],[@B3]\]. These verified cuticular protein sequences revealed motifs, unique to arthropod proteins, that have made it possible to classify sequences that came from cDNAs and genomes as cuticular proteins. In addition to sequence determination, studies of cuticular proteins have emphasized spatial distribution and expression in different developmental stages (reviewed in \[[@B2],[@B3]\]). Consequently, a wealth of information exists. We have used and organized this information in a relational database, named cuticleDB, the first database of arthropod cuticular proteins. The current total number of entries is 445, including proteins from 6 orders of Insects, 2 orders of Crustacea, and 2 orders of Chelicerata. This first version of cuticleDB is restricted to structural proteins of the cuticle; enzymes active in sclerotizing (tanning) or digesting cuticle and proteins involved in defense and pigmentation have been omitted. The database nomenclature is based either on the names given by those who deposited the sequences or on codes assigned by genome projects. Thus, we have retained the existing names/codes for the convenience of the users. Construction and content ======================== Data collection --------------- The data collection has been basically done in two ways. First, by submitting appropriate keywords (cuticle, exoskeletal, carapace) to the Protein databases of Entrez and Uniprot (release 1.8) \[[@B4]\] we collected a number of entries, which were manually filtered. Results from the two databases were checked to eliminate duplicates. Secondly, we obtained genome data for Anopheles gambiaeand Drosophila melanogaster, from Ensembl \[[@B5]\] and EBI, respectively. These are currently the only Arthropods with annotated genomes. We searched these genomes for a Pfam motif, PF00379, setting as cutoff the recommended gathering cutoff of the corresponding Pfam entry \[[@B6]\]. This motif has been shown to be responsible for chitin binding \[[@B7]\] and most probably adopts a precise, well-defined structure \[[@B8],[@B9]\]. A short version of this motif was first recognized by Rebers and Riddiford in 7 cuticular proteins \[[@B10]\], and, as more sequences became available \[[@B11]\], was widely recognized. The initial consensus was 35 amino acids long, but now encompasses 68 residues as sequence similarity was recognized at its amino-terminus and the carboxy-teminus was shortened. This 68 amino acid region, named the \"extended R&R consensus\" is what is recognized by PF00379 (\[[@B2],[@B3]\]and references therein). In order to ensure that our data collection is complete, we scanned all protein sequences of Uniprot (release 1.9) for PF00379. Again manual filtration was required. In addition to PF00379, other motifs have been described in cuticular proteins, some are found along with PF00379 while others define other families of cuticular proteins \[[@B2],[@B3]\]. All recognized cuticular protein motifs were used to construct the database. The data for our database was obtained by parsing the fields Definition, Accession, GI from Version, Organism and Origin from the Entrez entries. From the Uniprot entries we used Primary accession number, Protein name, Origin of the protein, Cross-references and Sequence information. This retrieval was done with Perl scripts. Additional information, concerning temporal and local expression of the proteins or corresponding mRNAs, was drawn from literature. Implementation -------------- The data have been organised based on a relational model and is stored in a PostgreSQL database system. The user has supervisory access through our Apache web-server. The database is managed by an interferential software, written in Java, which tends to settle any web-server\'s query. Also, it implements a homemade computational tool that performs motif search as described below. Data retrieval -------------- The main page of cuticleDB includes the following interfaces: Introduction, Data Retrieval, User manual and Contact. On clicking the Data Retrieval icon, users are presented with the search interface of the database. The query can be done in two ways: either by searching in fields or by gathering a set of proteins (Figure [1](#F1){ref-type="fig"}). The separate fields in which the user may search are Protein name, Taxonomy, references in other databases (the user may submit Entrez GenInfo Identifier, Entrez Accession Number, Uniprot AC, Flybase ID, Ensembl code, Interpro AC or Pfam AC as a query) and the protein sequence. The protein sequence can be searched against any pattern according to the user\'s imagination and, therefore, hypotheses for novel motifs can be tested. This is performed by a separate, homemade tool that has been integrated in cuticleDB and which gives the user the opportunity to detect new motifs in cuticular proteins. The integration of this tool is of importance especially in a database such as this, given the significance of motifs not only in cuticular proteins, but in structural proteins in general. Users can gather all protein entries from a single species (35 species are included in cuticleDB) or all protein entries whose protein sequence contains one of a series of motifs. However, this series of motifs has been pre-selected by the constructors of the database and cannot be modified by the user. The selection criterion was the frequency of appearance of these motifs in the literature. The most commonly found motifs were searched against all protein sequences of the database and have properly been assigned to each entry. Description of an entry ----------------------- A typical cuticleDB entry contains the following fields: Protein Name, References to other databases (Entrez Protein Database, Uniprot, Interpro, Pfam, Flybase, Ensembl), Taxonomy, Expression Details, Protein Sequence and its Length, Database-Source of the sequence and the method by which the sequence was obtained (Figure [2](#F2){ref-type="fig"}). The field \'Expression Details\' supplies the user with information about the anatomic region where each protein has been detected or the tissue where the corresponding mRNA is expressed, as well as the developmental stage in which the protein/mRNA appears. This field is usually accompanied by literature-citations. Moreover, another field named Patterns shows all patterns that have been searched for and found in the protein sequence, together with the start and end position of each. A text-box where the user can write his/her pattern is also available. If the user pattern matches the sequence, it is appended to the list of the predefined patterns. It remains there, as long as the user\'s session lasts. Also present are a field giving the known or predicted signal peptide and fields indicating whether the protein is putative, preliminary or fragment. Taxonomic distribution of the entries ------------------------------------- Taxonomic data are taken from Entrez. The total number of entries in cuticleDB is 445. These proteins are distributed in the three large taxa: Insecta (370 entries), Crustacea (60 entries) and Chelicerata (15 entries). The database includes entries from 6 orders of the class Insecta: Diptera (258 entries), Lepidoptera (39 entries), Orthoptera (37 entries), Hemiptera (6 entries), Coleoptera (22 entries), Dictyoptera (8 entries). The large number of proteins in Diptera is due to the inclusion of cuticular proteins from the two genomes (D. melanogaster, A. gambiae). The only verified cuticular proteins are those where the complete protein sequence or a unique N-terminal region was determined from a protein extracted from a cleaned cuticle or where a specific antibody reacted with proteins in cuticle or extracted from it. Finding mRNA in the epidermis is presumptive evidence that a protein is cuticular. The majority of cuticular proteins in this database were designated as cuticular proteins based on their sequence similarity to authentic cuticular proteins. Such proteins where sequence is the sole criterion for assignment are marked as \"putative\" in the database. Furthermore, at present, the annotation of the proteins of A. gambiaeis preliminary. Many proteins are missing signal peptides, other clearly have been incorrectly assembled. Such sequences are marked as preliminary as well as putative. This database will be continuously updated at regular intervals to accommodate annotation. The distribution of the proteins in the subphylum Crustacea is 59 entries from the order Decapoda, and 1 entry from the order Sessilia, whereas the distribution in the subphylum Chelicerata is 5 entries from the order Araneae and 10 from Xiphosura. Motif distribution ------------------ Apart from collecting and organizing data, this database also contains results of experimental computational work. Based on the classification of the \"extended R&R\" motif into two main types, RR1 and RR2 \[[@B12]\], which, at present, appears to correlate with their presence in proteins from soft and hard cuticles respectively, we built a Profile Hidden Markov Models for the two types. For this purpose we used the HMMER software package (Version 2.3.2) \[[@B13]\] utilizing its hmmbuild function. As an input to this function we used an alignment derived from 14 RR1 protein sequences from D. melanogasterfor the RR1 HMM and an alignment derived from 9 RR2 protein sequences from the same species for the RR2 HMM (suitably selected from reference \[[@B3]\]). Both of the alignments were restricted to the area of the \'extended R&R consensus\', thus they did not include the whole sequences. Subsequently, we used these Profile Hidden Markov Models as a prediction tool for classifying the cuticular proteins into two groups RR1, RR2. The prediction was in agreement with the literature as far as the known RR1 and RR2 proteins are concerned. The total number of RR1 and RR2 proteins in cuticleDB are 132 and 148, respectively. The start and end positions of the two motif-types are shown in the corresponding entry of each protein. A smaller class, RR-3, with 75 conserved residues was also identified by Andersen \[[@B14]\]. We have also studied the appearance of another motif: AAP(A/V). This small, hydrophobic tetrapeptide has been found to occur mainly in proteins of hard cuticles \[[@B2],[@B3]\], where the water content is low and the sclerotization is intense. We have found that the AAP(A/V) motif occurs in 43% of the RR2 proteins, whereas only in the 12% of the RR1 proteins of cuticleDB. Utility and discussion ====================== The most severe problem of genome projects to date is that of correct annotation. So, accurate and specialized databases as cuticleDB with its description of highly conserved motifs will be of help to genome annotators. Therefore, cuticleDB can be used as a basis for annotating new cuticular proteins by similarity in future Arthropod genome projects. cuticleDB can also be utilized in molecular research as well, due to its focus on motif appearance. Cuticular proteins, as is the case with all structural proteins are marked by the presence of characteristic motifs. Some motifs are repeated within a protein sequence, whereas others appear only once. cuticleDB has been designed in such a way that the user can have a complete view of motif occurrence in the sequence of each protein entry. First, each entry shows the exact position of the most common cuticle motifs in the protein sequence. Secondly, the user is given the opportunity to search the sequence for novel motifs and therefore, test hypotheses for the existence of new patterns. Subsequently, hypotheses for possible interactions between cuticle macromolecules (either proteins with chitin or proteins with proteins) can be tested. Moreover, our results of the RR1 and RR2 predictions can be used as a guide for identifying a certain protein as coming from either soft or from hard regions of the cuticle. Most importantly, the information about the RR1 and RR2 distinction can be used for studies of cuticle\'s mechanical properties. As RR1 and RR2 proteins appear in soft and hard cuticles respectively, which means that the former interact with chitin more loosely than the latter, one can gain an insight in cuticle\'s molecular construction combining our data on the sequences of RR1 and RR2 proteins with some experimental work. Moreover, one could use the Expression Details, namely where and when each protein is expressed, when studying the differential construction of the cuticle among different developmental stages or among different regions of a single cuticle. Conclusions =========== The goal of cuticleDB constructors was the collection of all cuticular protein sequences that have appeared to date and their detailed and correct annotation. The better the organisation of the data, the easier the work will be for researchers dealing with cuticle and structural proteins in general. cuticleDB will help them to answer questions like : \'What kind of proteins appear in hard cuticles?\' \'Why do RR2 proteins interact with chitin more tightly than RR1 proteins?\' \'Which motifs contribute to protein-protein interaction in the cuticle?\' \'From which stage can a certain protein be extracted?\' Furthermore, it is hoped that, detection of common properties of these proteins, as well as recognition of important differences that are responsible for cuticle\'s complexity and important functions will be facilitated by its existence. Last but not least, it is hoped that this database will be of help to genome annotators in the near future as more arthropod genomes become available. Availability and requirements ============================= cuticleDB was created and is maintained in the Department of Cell Biology and Biophysics, Faculty of Biology of the National and Kapodistrian University of Athens. It is freely available at the URL: <http://bioinformatics.biol.uoa.gr/cuticleDB>. An e.mail <biodb@biol.uoa.gr> may also be used for comments, corrections and further data (sequence) submission. List of abbreviations ===================== RR1: The extended Rebers and Riddiford Consensus, type I RR2: The extended Rebers and Riddiford Consensus, type II HMM: Hidden Markov Model Authors\' contributions ======================= CKM performed the data collection, and test procedures, and also participated in the design and the implementation of the database ICS carried out the design of the algorithms and the database,. implemented all the algorithms, and also created the web interface VAI supervised the data collection and the tests JHW compiled the first draft of known cuticular proteins, provided a critique of the data base during its construction SJH coordinated and supervised the whole project, suggesting the general directions and innovative features of the database All authors have read and accepted the final manuscript. Acknowledgements ================ We thank the University of Athens for financial support. This work was supported in part by grant AI055624 to JHW from the U.S. National Institutes of Health. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The Data Retrieval page of cuticleDB.The query can be done either by entering a word in the search fields Name, Taxonomy, Pattern and References or by gathering a set of entries that share a motif or derive from the same species. In this figure a query was made for all entries containing the word \'Arthropoda\' in their Taxonomy field. This happens to be one of the appropriate queries for getting all cuticleDB entries. ::: ![](1471-2105-5-138-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### A detailed view of a cuticleDB entry.This contains a number of fields: Protein Name, References to other databases, Taxonomy, Expression Details, Protein Sequence and its Length, Database-Source of the sequence and the method by which the sequence was obtained. The field Patterns shows all motifs found in the protein sequence, together with their start and end positions. Users can search for their own motifs, as well. The fields Signal peptide, Fragment, Putative and Comments follow. The entry of the figure corresponds to protein ACP20 from Tenebrio molitor. It was selected from the Result set, that appeared after the query with the word \'Arthropoda\'. ::: ![](1471-2105-5-138-2) :::	PubMed Central	2024-06-05T03:55:48.412408	2004-9-28	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522807/", "journal": "BMC Bioinformatics. 2004 Sep 28; 5:138", "authors": [ { "first": "Christiana K", "last": "Magkrioti" }, { "first": "Ioannis C", "last": "Spyropoulos" }, { "first": "Vassiliki A", "last": "Iconomidou" }, { "first": "Judith H", "last": "Willis" }, { "first": "Stavros J", "last": "Hamodrakas" } ] }
PMC522808	Background ========== The Adenomatous polyposis coli (APC) protein is an important tumour suppressor in the colonic epithelium \[[@B1]\]. A key function of this highly conserved protein is to antagonize Wnt signalling, by constitutively downregulating the transcriptional activity of β-catenin/Armadillo, a key effector of the Wnt signalling pathway \[[@B2]\]. Loss of this function is thought to be critical in the initiation of colorectal tumorigenesis as it causes a transcriptional switch in the intestinal epithelium towards actively dividing crypt progenitor cells \[[@B3]-[@B5]\]. APC proteins are highly conserved among vertebrates and flies, and flies encode two APC proteins with overlapping roles in Wnt signalling during development \[[@B6],[@B7]\]. However, APC proteins have additional functions in connection with the actin and microtubule cytoskeletons that appear to be separate from their function in controlling Wnt signalling \[[@B8],[@B9]\]. One of these functions is a role of APC in facilitating cellular adhesion, as indicated by studies in Drosophila tissues \[[@B10]\] and in mammalian colorectal cancer cells \[[@B11]\]. This function in cellular adhesion is likely to be conferred by the subcellular pool of APC protein that is associated with adherens junctions (AJs) in Drosophila \[[@B12],[@B13]\] and in polarised mammalian cells \[[@B14]\]. The mechanism by which APC facilitates cellular adhesion is unknown. In order to explore this mechanism, we asked whether Drosophila E-APC (also called dAPC2) might have a structural role at AJs. If so, E-APC would be expected to be stably associated with AJs, similarly to the structural components of the adhesive complex. As in mammalian epithelia \[[@B15],[@B16]\], the main functional components of this complex in Drosophila are the transmembrane protein E-cadherin, a calcium-dependent trans-membrane adhesion molecule, and the catenins (Armadillo and α-catenin) that link E-cadherin to the actin cytoskeleton at the cytoplasmic side \[[@B17]-[@B22]\]. We thus conducted photobleaching experiments on live embryos expressing E-APC or structural AJ components tagged with green fluorescent protein (GFP) \[[@B23]-[@B25]\], to compare their relative mobility. These experiments revealed that GFP-E-APC is less stably associated with AJs than their structural components. We also found that GFP-E-APC is remarkably mobile in neurons. Results and discussion ====================== We used the GAL4 system to express GFP-E-APC throughout the embryo, and found that its subcellular distribution is very similar to that of endogenous E-APC in fixed embryos. In particular, GFP-E-APC is concentrated underneath the plasma membrane in apicolateral regions of embryonic epithelial cells (Fig. [1a](#F1){ref-type="fig"}; Fig. [2b](#F2){ref-type="fig"}). These regions form the zonula adherens (ZA) which contains the AJs \[[@B26]\]; they can also be visualised with antibody staining against α-catenin \[[@B27]\] (Fig. [1b](#F1){ref-type="fig"}), and we observe a remarkably close coincidence of GFP-E-APC and α-catenin. Similar results were obtained by Akong at el. who examined the same GFP-E-APC transgene in the embryo \[[@B7]\], and who also showed that GFP-E-APC is distributed similarly as endogenous E-APC in larval neuroblasts \[[@B28]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Association of GFP-E-APC with AJs of embryonic epithelial cells. Side view of epidermis of a \~6 hour old embryo expression ubiquitous GFP-E-APC, stained with antibody against (a) GFP and (b) α-catenin to mark the apicolateral AJs; note the co-incidence of GFP-E-APC and α-catenin staining (arrows). ::: ![](1471-2121-5-37-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### FRAP protocol for the analysis of live Drosophila embryos expressing GFP-E-APC. (a) Sketch of the microscopy and data analysis used to determine the mobility of GFP-E-APC in epithelial cells of live embryos. (b) Consecutive face-on views of a live \~6 hours old embryo (stage 11) expressing GFP-E-APC, with squares marking specific sections of the AJs at cell interfaces; the set of yellow squares in the right-hand image illustrate the cell shape changes that took place during the 5 minutes between the two optical sections shown. White bar in this and subsequent figures, 5 μm. ::: ![](1471-2121-5-37-2) ::: Next, we conducted fluorescence recovery after photobleaching (FRAP) experiments in live embryos, to examine how stably GFP-E-APC is associated with adherens junctions. We bleached the fluorescence in a defined square centred over the junctional region of an epithelial cell with a short laser pulse, and examined the recovery of the fluorescence within this square over time (Fig. [2a](#F2){ref-type="fig"}). This revealed a relatively fast rate of recovery of most of the fluorescence within a few minutes (Fig. [3a](#F3){ref-type="fig"}) \[see [additional file 1](#S1){ref-type="supplementary-material"}\]. Quantitative analysis shows that nearly 80% of the initial fluorescence is recovered within \~220 seconds (with a half-time value of \~60 seconds) (Fig. [4a,4e](#F4){ref-type="fig"}). This indicates that the association of E-APC with the ZA is dynamic rather than stable. The fluorescence recovery observed in these FRAP experiments could be due to movement of E-APC between the cytoplasm and the plasma membrane, but also to sideways movement along the plasma membrane (see also below). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### FRAP of GFP-E-APC and GFP-tagged AJ proteins in early embryonic epithelial cells. Face-on views of live \~6 hours old embryos (stage 11) expressing (a) GFP-E-APC, (b) E-cadherin-GFP, (c) Armadillo-GFP, (d) α-catenin-GFP, with white squares marking sections of cell interfaces that were bleached, and red squares marking unbleached control sections. Pre-bleaching images are shown on the left; subsequent images on the right show recovery of fluorescence within white squares 15, 75 and 300 seconds after bleaching \[see additional file 1\]. ::: ![](1471-2121-5-37-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Quantitative evaluations of FRAP experiments. Plots of the relative fluorescence in white squares compared to grey squares in Fig. 3 as a function of time; error bars indicate the range of values from \>12 different experiments. (a-d) FRAP of individual GFP-tagged proteins in wild-type embryos, as indicated; (e) combined data of (a-d); (f) Comparison of FRAP of GFP-E-APC in wild-type and sggmutant embryos (note that we were unable to distinguish between null mutant and paternally rescued embryos; however, the rescue activity of the paternal allele is minimal as both types of embryos are highly abnormal). ::: ![](1471-2121-5-37-4) ::: We also conducted FRAP experiments with structural AJ components, namely E-cadherin-GFP, Armadillo-GFP and α-catenin-GFP. In these cases, we can only recover a small fraction of the initial fluorescence within the time frame of the experiment (Fig. [4a,4b,4c,4d,4e](#F4){ref-type="fig"}; note that these experiments cannot be extended beyond \~6 minutes, due to the extensive cell shape changes during this developmental stage). Furthermore, the rate of recovery is slower than that observed with GFP-E-APC, with estimated half-times of \>3 minutes (α-catenin-GFP and of E-cadherin-GFP; Fig. [4b,4d,4e](#F4){ref-type="fig"}). This also appears to be true for Armadillo-GFP (Fig. [4c,4e](#F4){ref-type="fig"}), though we cannot estimate its half-time of recovery with confidence, given that its fluorescence levels are considerably lower than that of the other GFP-tagged protein examined in this study. We conclude that E-APC is significantly more mobile than the structural AJ components. This suggests that E-APC shuttles either within the cortex, along the zonula adherens, or that it shuttles from the cytoplasm to the plasma membrane (as previously proposed; \[[@B29]\]). Interestingly, the observed rates of recovery of GFP-E-APC were much slower than the estimated rate of free diffusion (e.g. \[[@B30]\]; the rate of recovery of GFP alone was \<10 seconds, i.e. too fast to be measured by our experimental setup). This suggests that the movements of GFP-E-APC are primarily determined by the kinetics of its binding to ligands. One of these could be Axin which associates with E-APC in Drosophila cells to from large dot-like structures \[[@B31]\]. Similarly, Axin associates with APC in mammalian cells to form large molecular weight protein complexes \[[@B32]\]. Our observations argue against a structural role of E-APC in cellular adhesion. However, they are consistent with a catalytic role of E-APC in facilitating cellular adhesion, for example by maintaining the junctional pool of Armadillo \[[@B10],[@B29],[@B33]\]. In support of this, recent evidence suggests that there is rapid exchange of β-catenin within the junctional cadherin complex, and that APC is required for this process \[[@B34]\]. In late embryonic stages, GFP-E-APC forms striking patches underneath the apical plasma membrane of epidermal cells that are in the process of forming denticle extrusions (Fig. [5](#F5){ref-type="fig"}). These striking \'pre-denticle\' patches are also seen in embryos stained with antibody against E-APC, and overlap with actin patches \[[@B25]\]. They may thus represent an actin-dependent association of E-APC as that seen in the cortex of earlier epithelial cells and at the ring canals between nurse cells within the egg chambers \[[@B25],[@B33]\]. FRAP experiments revealed that the presence of GFP-E-APC in these pre-denticle patches is also dynamic, with an estimated half-time of fluorescence recovery of 200--300 seconds (Fig. [5](#F5){ref-type="fig"}). Again, E-APC is therefore unlikely to have a structural role in these patches. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### FRAP of GFP-E-APC patches in late embryonic epithelial cells. Face-on views of \~17 hours old embryo (stage 17) expressing GFP-E-APC, showing patches of GFP-E-APC at the apical plasma membrane of epidermal cells forming denticles (pre-bleach and subsequent images labelled as in Fig. 3). Note the fast recovery of the fluorescence in these patches after photobleaching. ::: ![](1471-2121-5-37-5) ::: It has been reported that E-APC and Armadillo are required for anchoring mitotic spindles in the cortex of dividing blastoderm cells in the early Drosophila embryo \[[@B25]\]. We cannot measure the kinetics of GFP-E-APC association with the cortex in these early embryonic cells, because of insufficient expression levels at this stage. However, assuming that these kinetics do not change radically during embryonic development, our observations from the later embryos (Fig. [3](#F3){ref-type="fig"},[4](#F4){ref-type="fig"},[5](#F5){ref-type="fig"}) suggest that E-APC has a catalytic role in capturing microtubules in the cellular cortex, rather than providing a structural tether \[[@B25]\]. We also expressed GFP-E-APC in eye imaginal discs, to examine its subcellular distribution within a larval epithelial sheet. We thus noticed striking puncta of green fluorescence within the axons of the optic stalk that connects these discs to the larval brain (Fig. [6](#F6){ref-type="fig"}). These puncta resemble the E-APC/Axin-GFP dots that we observe in embryonic cells \[[@B31]\] and in these axons (not shown), and also the E-APC/Armadillo dots that Peifer and colleagues observed in early embryos \[[@B25]\]. They may thus represent the Axin destruction complex \[[@B31]\]. We performed FRAP experiments, bleaching a 6 μm wide strip perpendicularly across the axons and monitoring the recovery of the fluorescence into the bleached section. This revealed that the GFP-E-APC puncta are remarkably dynamic: they re-appear within the bleached area within a minute, with an estimated half-time of \~100 seconds (Fig. [6](#F6){ref-type="fig"}) \[see [additional file 2](#S2){ref-type="supplementary-material"}\]. Many of these puncta seem to re-appear from other focal planes, so we cannot be absolutely certain that they represent movement of existing puncta. However, some of the puncta re-appearing in the bleached area can clearly be traced as moving puncta within the same focal plane (e.g. see isolated axon, left-hand side of \[[additional file 2](#S2){ref-type="supplementary-material"}\]). The movement of these GFP-E-APC puncta may be due to tracking (e.g. along microtubules), although we cannot see uni-directionality of movement (i.e. the movement appears to be up and down the axons). The movement we observe in these axons is somewhat reminiscent of that observed with a GFP-tagged truncation of Xenopus APC that misses its C-terminus (thus resembling the overall structure of E-APC): this truncation, despite lacking the putative microtubule-interacting domain within its C-terminus, forms large puncta that can track along microtubules in Xenopus tissue culture cells \[[@B35]\]. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### FRAP of GFP-E-APC in the larval optical stalk. Optical sections through the optical stalk of a third instar larva, before and after photobleaching, showing fluorescent puncta in individual axons, and the reappareance of these puncta (arrows) from both sides of the bleached areas within minutes \[see additional file 2\]. Width of bleached strip, 6 μm. ::: ![](1471-2121-5-37-6) ::: GSK3 is inhibited during Wnt signalling \[[@B2]\], and GSK3 mutants in Drosophila (shaggy/zeste white3, or sgg, mutants) therefore mimic constitutive and sustained Wingless signalling \[[@B36]\]. The normal level of Wingless signalling in the embryonic epidermis does not appear to change the subcellular distribution of bulk E-APC protein \[[@B12]\], although it does cause a re-location of Axin-GFP/E-APC complexes to the plasma membrane \[[@B31]\]. However, a reduction of cortical E-APC has been observed in early sggmutant embryos \[[@B12],[@B25]\]. Likewise, in older sggmutant embryos, the levels of membrane-associated GFP-E-APC are also noticeably reduced (Fig. [7a,7b,7c](#F7){ref-type="fig"}). However, this does not appear to be due to a change in mobility of GFP-E-APC since the kinetics of fluorescence recovery between wild-type and sggnull mutant embryos were comparable (Fig. [4f](#F4){ref-type="fig"}). Instead, it is due to a reduction of the overall E-APC protein levels in these mutants: Western blot analysis of 2--16 hour old embryos revealed that the total levels of GFP-E-APC protein were much lower in sggmutant embryos compared to the wild type (Fig. [7d](#F7){ref-type="fig"}). The same is true for endogenous E-APC whose levels are also substantially reduced in sggmutants (Fig. [7e](#F7){ref-type="fig"}). This indicates that sggis required for the stability of E-APC protein, and it suggests that sustained Wingless signalling may destabilise E-APC. Similarly, phosphorylation by GSK3 is required for the stability of mammalian Axin, a functional binding partner of APC \[[@B37]\], and the levels of Drosophila Axin in embryos are also reduced after prolonged Wingless signalling \[[@B38]\]. Destabilisation of the main components of the Axin complex (Axin and APC) during Wnt signalling may be a positive feedback mechanism resulting in the amplification of the signalling level. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Destabilisation of E-APC in sggmutant embryos. (a, b) Face-on and (c) side views of \~14 hours old embryos (stage 16), fixed and co-stained with antibodies against GFP and α-catenin as indicated, revealing junctional association of GFP-E-APC in (a) wild-type and (b, c) sggmutant embryos (similar in sggnull and paternally rescued embryos, see also Fig. 4f). (d, e) Western blots of hand-picked 10--16 hours old wild-type and sggmutant embryos (\~100 embryos per lane), probed with antibodies against (d) GFP or (e) E-APC, and α-tubulin as internal controls. Note that the levels of GFP-E-APC and of endogenous E-APC are much reduced in sggcompared to wild-type embryos (sggmutants represent a 1:1 mixture of sggnull and paternally rescued embryos). The lower bands in upper panels (d, e) correspond to breakdown products of GFP-E-APC and E-APC, respectively; their occurrence varies somewhat between preparations. ::: ![](1471-2121-5-37-7) ::: The subcellular distribution of E-APC and its accumulation at the adherens junctions is unchanged in other mutants of the Wingless signalling pathway (including wg, axin, dshand signalling-defective armmutants; \[[@B31]\]; F. Hamada, X. Yu and M. B., unpublished observations). We thus did not expect any of these mutants to affect the shuttling behaviour of GFP-E-APC to and from the plasma membrane. In support of this, preliminary FRAP experiments indicated that the kinetics of fluorescence recovery are unaffected in dshnull mutant embryos (not shown). Taken together with our results from the sggmutants, this suggests that the kinetic association of GFP-E-APC with the plasma membrane is unaffected by Wingless signalling. Conclusion ========== Our FRAP experiments provided evidence that E-APC is a cytoplasmic shuttling protein whose association with the adherens junctions is highly dynamic. The speed of its shuttling to and from the plasma membrane appears to be constitutive and does not require GSK3 activity. The dynamic association of E-APC with the plasma membrane is consistent with a catalytic role of E-APC, and argues against a structural or tethering role in the cell cortex. Methods ======= Fly strains ----------- Fly lines transformed with UAS.GFP-E-APC (full length E-APC tagged with GFP at its N-terminal end, inserted into pUAST \[[@B39]\]) were generated by R. Rosin-Arbesfeld (see also \[[@B7],[@B28]\]). The GAL4 driver lines arm.GAL4 and GMR.GAL4 (FlyBase) were used to express GFP-E-APC throughout the embryonic epidermis \[[@B31]\] and in the larval eye disc, respectively. All fly strains were cultured at 25°C. zw3^M11-1^and dsh^v26^mutant embryos lacking maternal and zygotic gene function were generated as described \[[@B40]\]. We did not detect any differences in the subcellular localisation of GFP-E-APC or α-catenin between zygotic null and paternally rescued sggmutants (identified with an RFP-marked X chromosome \[[@B41]\]). For Western blot analysis, 10--16 hours old wild-type and sggmutant embryos were hand-picked (from timed egg collections) under the dissecting microscope, and separated into GFP-positive and GFP-negative embryos; unfertilised embryos were discarded. Analysis of fixed embryos and Western blots ------------------------------------------- Antibody staining of fixed embryos and analysis by confocal microscopy were described previously \[[@B12]\]. The following primary and secondary antibodies were used: rabbit anti-E-APC \[[@B12]\], rabbit anti-GFP \[[@B14]\], rat anti-α-catenin \[[@B42]\]; goat anti-rat IgG Alexa Fluor 568, goat anti-rabbit IgG Alexa Fluor 488 (Molecular Probes). The following primary and secondary antibodies were used for Western blotting: rabbit anti-E-APC \[[@B12]\]; mouse monoclonal anti-GFP IgG2a (Santa Cruz Biotechnology); mouse anti-α-tubulin (clone B-5-1-2, Sigma), as internal control; goat anti-mouse and anti-rabbit HRP IgG (Santa Cruz Biotechnology). The enhanced chemiluminescence (ECL) Western blotting system (Amersham) was used for detection \[[@B43]\]. Live imaging of embryos ----------------------- For live imaging, embryos were dechorionated in 50% bleach for 1--2 minutes and washed. Embryos were transferred to a moistened black filter (Schleicher and Schüll). Embryos were adhered to coverslips with heptane glue, made by mixing heptane and clear sellotape (Sellotape Ltd). Embryos were mounted in Voltalef oil (10S). For short term imaging (\<30 minutes), embryos were mounted on a glass slide with small coverslips as supports. For longer term imaging, e.g. for bleaching of pre-denticle patches, embryos were mounted in oil and placed on Bio-foil gas permeable membrane (Sartorius Ltd) mounted on a perspex frame \[[@B44]\]. Photobleaching of live embryos ------------------------------ FRAP experiments were performed using a Bio-Rad Radiance confocal microscope with a 40× NA 1.3 objective lense. Imaging was performed with a 488 nm argon laser at 5% laser power and the following confocal settings: iris at 4 mm, 50% gain, zoom 10, scan speed 500 lps, box size 512 × 512 pixels. These conditions were found to give minimal photobleaching over the observed time. For each FRAP experiment, a pre-bleach image was recorded by selecting a focal plane and taking a Z-series, consisting of 3 0.5 μm steps either side of the desired focal plane (from -1.5 μm to +1.5 μm). The LaserSharp software was used to define several regions of interest (ROIs) for bleaching. A maximum of one bleach ROI was placed in any cell and several cells were always left unbleached. Typically, 3--5 ROIs were bleached in one field of view on one embryo. These regions were bleached at 100% laser power (scanning at 500 lps). 10 bleach scans were found to produce the best results for all constructs. After bleaching, a Z-series was recorded every 15 seconds for 5 minutes. At the time of these experiments, the LaserSharp software did not contain a function for performing this type of 4D bleaching experiment. This problem was overcome by manually switching between imaging and bleaching settings and manually saving pre bleach images and starting the time course. As a result of this, there was usually a 30--60 second delay between the pre-bleach image and the post-bleach images. Data analysis ------------- Data sets were analysed with the Bio-Rad LaserPix software. For each time point, the total pixel intensity distribution was compared to the pre-bleach image to select the corresponding region. The two images were then compared by eye to confirm that they did correspond to the same focal plane. The coordinates for the bleach ROIs were used to accurately locate the bleach spots on the pre bleach image, and the mean fluorescence intensity for each ROI was calculated. Several equivalent sized ROIs were also placed on unbleached cells to measure any change in fluorescence due to photobleaching or movement. To track movement of the cells, an acetate sheet was placed over the computer monitor and each ROI was marked on it as well as the shapes of the cells surrounding it. By aligning the sheet with the appropriate cell shapes, the ROI could be appropriately positioned for each time point. This process was used to position each ROI on the appropriate image for each time point. Once all ROIs had been placed on the image, the mean fluorescence intensities were calculated for each ROI, and their positions were saved on a copy of the image (See Fig [2.1](#F2){ref-type="fig"}). Data was exported to Microsoft Excel for analysis. Relative fluorescence was calculated for each bleach area by dividing fluorescence at time (t) by pre-bleach fluorescence. The change in fluorescence was plotted on a graph with Excel. For each construct tested, the data from multiple bleach experiments from multiple embryos were averaged to give the approximate rate of recovery. Data sets were discarded for any of the following reasons. First, if movement of the embryo in the Z axis took the sample outside the range of the Z-series in any time point. Second, if movement in the X/Y axis was sufficient to move significant numbers of the bleach boxes outside of the observed region. Third, if an ROI ever left the field of view, all data points for that ROI was discarded. Fourth, all data sets were discarded if the intensities of the control ROIs changed dramatically at any point in the experiment, or showed a large general increase or decrease. Pre-denticle structures were bleached in a similar manner to junctional E-APC described above. Live imaging of the larval optic stalk -------------------------------------- GFP-E-APC was expressed in eye imaginal discs by the GAL4 system, using the driver line GMR.GAL4 (described in FlyBase). Eye discs and brains were dissected from crawling third instar larvae in PBS. Eye discs were teased away from the brain and inverted to reveal the optic stalk. Whole disc/brains were mounted in a drop of PBS under a cover slip, supported by two smaller cover slips. Each disc was observed for no more than 30 minutes. Photobleaching of the larval optic stalk ---------------------------------------- FRAP experiments were performed using a Bio-Rad Radiance confocal microscope and Bio-Rad LaserSharp software, using the 100× NA 1.4 objective lens. A narrow strip was bleached across the whole field of view by adjusting the size of the scanning area. These experiments were performed before a FRAP program was available for LaserSharp so bleaching was performed manually, leading to somewhat variable intervals between each stage of the experiment. The region was bleached with the 488 nm line of an argon laser for approximately 20 scans. Time courses were recorded after each bleaching experiment for 5 minutes. Authors\' contributions ======================= A.C. developed and conducted most of the FRAP experiments; J.M. completed the analysis of GFP-E-APC in live and fixed sggmutant embryos, including the Western blots; M.B. directed the study, helped with the microscopy and drafted the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 FRAP of GFP-E-APC in early embryonic epithelial cells. Example of a FRAP experiment of GFP-E-APC, as described in Figure 3. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 FRAP of GFP-E-APC in the larval optical stalk. Example of a FRAP experiment of GFP-E-APC, as described in Figure 6. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank R. Rosin-Arbesfeld, H. Oda and M. Peifer for fly strains expressing GFP-tagged proteins, and F. Hamada for help with the Western blots. J. M. is supported by a grant from the Association for International Cancer Research (no. 03--275) awarded to M.B.	PubMed Central	2024-06-05T03:55:48.414382	2004-9-30	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522808/", "journal": "BMC Cell Biol. 2004 Sep 30; 5:37", "authors": [ { "first": "Adam", "last": "Cliffe" }, { "first": "Julius", "last": "Mieszczanek" }, { "first": "Mariann", "last": "Bienz" } ] }
PMC522809	Background ========== Present-day protein sequences can be used to reconstruct ancestral sequences based on a model of sequence evolution. Such knowledge about ancestral sequences is helpful for understanding the evolutionary processes as well as the functional aspects of a protein family. Existing methods of ancestral sequence reconstruction can be divided into two main categories: Maximum Parsimony (MP) methods \[[@B1],[@B2]\] and Maximum Likelihood (ML) methods \[[@B3]-[@B5]\]. MP methods do not take into account biased substitution patterns between amino acids or different tree branch lengths, and cannot distinguish those equally parsimonious reconstructions \[[@B3]\]. ML methods do not have these limitations and generally give more reliable results than the MP methods \[[@B6]\]. Yang et al. \[[@B3]\] first developed a ML method for ancestral sequence reconstruction. Yang \[[@B7]\] also made a distinction between \"joint\" reconstruction and \"marginal\" reconstruction. Joint reconstruction methods intend to find the most likely set of amino acids for all internal nodes at a site, which yields the maximum joint likelihood of the tree \[[@B5]\]. Marginal reconstruction compares the probabilities of different amino acids at an internal node at a site and selects the amino acid that yields the maximum likelihood for the tree at that site. Marginal reconstruction can also compute probabilities of all other amino acids for that node \[[@B4]\]. Koshi and Goldstein \[[@B4]\] developed a fast dynamic programming algorithm for marginal reconstruction in the framework of Bayesian statistics, while Pupko et al. \[[@B5]\] proposed a fast algorithm for joint reconstruction. The computational complexities for both algorithms scale linearly with the number of sequences. Both marginal and joint reconstruction algorithms are implemented in our program. All reconstruction methods require a phylogenetic tree inferred from a given alignment. The quality of the tree is crucial for the reliability of reconstruction. A number of methods exist for phylogenetic inference, such as maximum likelihood \[[@B8]\], distance-based \[[@B9]\] and parsimony \[[@B1]\]. Distance-based methods have the advantage of being simple and are able to handle a large set of sequences. They require evolutionary distances estimated for all the sequence pairs. The most common method to infer phylogeny from distances is based on the neighbor-joining algorithm \[[@B9]\]. Bruno et al. \[[@B10]\] introduced a distance-based phylogeny reconstruction method called \"Weighbor\", i.e. \"weighted neighbor joining\", which takes into account the fact that errors in distance estimates are larger for longer distances. Giving similar results, Weighbor is much faster than ML phylogeny reconstruction. It is also better than other methods such as BIONJ \[[@B11]\] and parsimony \[[@B1]\], in aspects of \"long branches attract\" and \"long branch distracts\" problems \[[@B10]\]. Weighbor is used in our program for phylogenetic inference. Overwhelming evidence exists for substitution rate variation across sites \[[@B12]-[@B15]\]. For a protein family, rate heterogeneity reflects the selective pressure imposed by folding, stability and function. Gamma distribution is widely used to model the rate variation among sites \[[@B13],[@B16],[@B17]\] because of its simplicity. Nielsen \[[@B18]\] suggested a method for site-by-site estimation of rate factors by a Maximum Likelihood approach. Rate variation among sites has not been taken into account in the early work of ML reconstruction of ancestral sequences \[[@B4],[@B5]\]. Recently, Pupko et al. \[[@B19]\] introduced rate variation into joint reconstruction by a branch-and-bound algorithm, assuming a gamma distribution of rates among sites. In our package, two methods are proposed to estimate a rate factor for each site. The first one is based on our observation that the substitution rate at a site is correlated with the conservation of the site. The more conserved the site is in a multiple sequence alignment, the smaller its substitution rate is. This empirical method, the result of which we call Alignment-Based rate factors or α~AB~, relies only on a multiple sequence alignment and a general model of amino acid exchange. The other one is a maximum likelihood method (α~ML~), which requires a tree. In our implementation, we incorporate α~AB~or α~ML~in the joint and marginal reconstruction algorithms \[[@B4],[@B5]\]. α~AB~is also used in the Maximum Likelihood estimation of evolutionary distances \[[@B20]\] for tree inference. We implement a method of evolutionary simulation that introduces site-specific rate variations in a natural way by imposing structural and functional constraints \[[@B21]\]. We show by simulations that the reconstruction methods can give reasonable results and that the problem of evolutionary distance underestimation \[[@B22]\] is alleviated by considering rate variation across sites. Background (or equilibrium) amino acid frequencies (π) are usually estimated from the target set of sequences or from large databases of protein families. Background amino acid frequencies estimated from a small dataset tend to have bias, while amino acid frequencies from large databases may not be suitable for the specific protein family under analysis. Here, we propose a ML method to optimize the amino acid frequency vector π. The optimized πvector can give significant improvement over the likelihood of a alignment. Information obtained from ancestral sequence reconstruction is used for two applications: homology detection and prediction of functional sites. For homology detection, ancestral sequences represent an enlargement of the sequence space around native sequences. We demonstrate that adding reconstructed ancestral sequences to a native alignment improves the detection of homologs in database searches. A number of methods have been developed to predict functional sites from amino acid sequences \[[@B23],[@B24]\]. One simple way to infer functional sites is by positional conservation of a multiple sequence alignment \[[@B25]\]. Lichtarge et al. \[[@B26]\] proposed a method called evolutionary trace to predict functional sites by analyzing the conservation of sequence subgroups. Functional divergence during the evolutionary process can be reflected in the variation of amino acid usage across different functional subgroups. We propose a new approach that uses information from ancestral sequence reconstruction to identify sites that are well conserved within individual sub-trees but exhibit variability among different sub-trees. By several examples, we show that these sites frequently contribute to the functional specificity of a protein family. Results and discussion ====================== We developed a package (ANCESCON) to reconstruct ancestral protein sequences considering rate variation among sites. Rate factors can be estimated either by an empirical method or by a maximum likelihood method. Consideration of rate variation among sites not only improves evolutionary distance estimation, but also gives more accurate ancestral sequence reconstruction. Ancestral sequences are used to improve profile-based sequence similarity searches. We also propose a new approach to predict positions with functional specificity based on the reconstruction of ancestral sequences. Observed α, Alignment Based Rate Factor α(α~AB~) and Rate Factor αestimated by Maximum Likelihood (α~ML~) ----------------------------------------------------------------------------------------------------------------------- Evolutionary simulations based on a Z-score model introduce rate variation across sites in a natural way by incorporating structural and functional constraints specific for a protein family \[[@B21]\]. The simulation procedure is a Monte Carlo simulation of the amino acid substitution process. The fixation of substitutions is dictated by a simple scoring function, which is derived from the template structure and an alignment of its homologs. The number of substitutions occurring at each site can be recorded during the simulation process and the observed αat a site equals the number of recorded substitutions at that site divided by the average substitution number for all sites. To reduce sampling variance, an average observed αvector is calculated from 100 simulations. For the alignment consisting of all the leaf node sequences generated by the simulation process, an α~AB~vector was calculated according to equation (11) (for details see Methods). An average α~AB~vector was derived from 100 simulations. Correlation coefficient between the average α~AB~vector and the average observed αvector was high (data not shown). However, we found that for large observed αvalues, the corresponding α~AB~values were smaller. A constant βwas introduced to correct this underestimation in equation (11). ![](1471-2148-4-33-i1.gif) Here, α~i~is Alignment-Based rate factor at site i. Kis the number of sites in a given alignment. C~i~is the value assigned to site i(for details see Methods). We optimized the βvalue by fitting the average α~AB~vector and average observed αvector to y= xline. Alignments for three different protein families (trypsin, carboxypeptidase and pdz domain) gave a good empirical estimation for βof about 1.3. The relation between this corrected average α~AB~vector and average observed αvector is shown in Figure [1a](#F1){ref-type="fig"} for a typical example, the pdz domain (correlation coefficient 0.973). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### *a) Correlation between average α~AB~and average observed α. b) Correlation between average α~ML~and average observed α.*α~AB~is Alignment-Based rate factor solely depending on the given alignment. α~ML~is rate factor estimated by maximum likelihood method, which requires an alignment and evolutionary tree inferred from the alignment. The protein family used here is the PDZ domain. ::: ![](1471-2148-4-33-1) ::: We also estimated an α~ML~vector for each alignment generated from the simulation (for details see Methods). The average α~ML~vector shows good correlation with the average observed αvector (Figure [1b](#F1){ref-type="fig"}) (correlation coefficient 0.945). α~AB~or α~ML~can be incorporated in likelihood calculation in marginal or joint reconstruction. Table [1](#T1){ref-type="table"} shows that improvement of logarithm likelihood of the alignment is significant when α~AB~or α~ML~is used. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Difference of logarithm likelihood and CPU time when using different αvectors ::: α= 1.0 α~AB~ α~ML~ ---------------------- ---------- ----------- ----------- ---------- ---------- -------- ---------- Δl P\* Δl P\* Logarithm Likelihood -5324.56 -5087.72 236.84 \<0.0001 -4987.27 337.29 \<0.0001 CPU Time (s)^+^ 213 213 359 The alignment tested here is a subset of SH2 family. It includes 44 sequences and each sequence contains 83 amino acids (including gaps). \* The likelihood ratio test (LRT) \[58\] is used to test whether α~AB~and α~ML~are significantly different from α= 1.0. The difference in number of free parameters between α~AB~, α~ML~and α= 1.0 model is 82. ^+^CPU times were computed on a Dell PowerEdge 8450 server (CPU 700MHz, RAM 8G). ::: Rate variation across sites can be modeled by assuming that the rate factors follow a certain type of statistical distribution. Gamma distribution \[[@B13],[@B27]\] and its discrete approximations \[[@B28]\] are frequently used for DNA or protein sequences. Rate variation for a protein family reflects different selective pressure at different sites to maintain structure and function. Fewer substitutions are expected to occur in more conserved sites. This hypothesis has prompted us to estimate rate factors (α~AB~) based on sequence conservation in an empirical way. The α~AB~is compared and calibrated using the observed αas standards. Our method of estimating α~ML~is similar to the one proposed by Nielson \[[@B18]\]. One problem with site-by-site rate factor estimation is the small sample size at each site, especially with a small alignment. We have used α~AB~to eliminate outliers with very large α~ML~estimates (for details see Methods). Site-specific rate factors improve distance estimation ------------------------------------------------------ Evolutionary distances tend to be underestimated when rate homogeneity among sites is assumed \[[@B22]\]. This was tested using the simulation with structural and functional constraints. For the arbitrarily selected tree shown in Figure [2](#F2){ref-type="fig"}, we obtained leaf node sequences in the simulation and estimated an evolutionary distance for each sequence pair by Maximum Likelihood, either incorporating α~AB~or setting αequal to 1.0 (equation (16)). Evolutionary distances were severely underestimated (average underestimation: 0.894) without considering rate variation among sites (Figure [3a](#F3){ref-type="fig"}). Introducing α~AB~in the maximum likelihood method gave more accurate distance estimation (Figure [3b](#F3){ref-type="fig"}), although the distances were still underestimated, especially for small distances (average underestimation: 0.286). We believe that more accurate distances will give more accurate phylogeny reconstruction using \"Weighbor\" \[[@B10]\]. Since a tree is required to estimate α~ML~, α~ML~is not incorporated in estimating evolutionary distance. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### The tree used to test ancestral sequence reconstruction.This is an arbitrarily selected evolutionary tree. Evolutionary distances are shown to scale. ::: ![](1471-2148-4-33-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### *Comparison of pairwise distances between the rebuilt tree and original tree. a) distance estimation assuming no rate variation among sites; b) distance estimation with α~AB~.The rebuilt tree is inferred from the alignment that is generated by evolutionary simulation performed on the original tree. The original tree is arbitrarily selected. ::: ![](1471-2148-4-33-3) ::: Optimization of equilibrium frequencies --------------------------------------- A continuous minimization method by simulated annealing was used to optimize the equilibrium frequency vector π, with the objective function being the logarithm likelihood of the alignment. Our πvector optimization program was tested on four alignments, which were taken from the SH2 and SH3 superfamilies in Pfam database (version 7.3) \[[@B29]\]. Two alignments from the SH2 superfamily have 44 and 87 sequences respectively and both alignment lengths are 83 amino acids (including gaps). The other two alignments from SH3 superfamily have 39 and 94 sequences respectively and both alignment lengths are 57 amino acids (including gaps). For each alignment, we ran optimization 3 times starting from different random initial points. The optimized πvectors did not converge to exactly the same point, but they had a high correlation with each other (always \> 0.95) and the difference of logarithm likelihood function values was small (less than 0.1%). The logarithm likelihood of the alignment, using optimized πvector, increased slightly, but significantly (Table [2](#T2){ref-type="table"}), compared with the logarithm likelihood using the πvector calculated from the alignment. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Difference of logarithm likelihood and CPU time with and without optimization of πvector ::: α~AB~& Calculated π* α~AB~& Optimized π Δl P\* ---------------------- --------------------------- -------------------------- ------- ---------- Logarithm Likelihood -5087.72 -5055.97 31.75 \<0.0001 CPU Time (s)^+^ 213 14902 The alignment tested here is the same alignment used in Table 1. Calculated πmeans frequency vector calculated from the alignment. \* The likelihood ratio test (LRT) \[58\] is used to test whether optimized πis significantly different from calculated π. The difference in number of free parameters between these two models is 19. ^+^CPU times were computed on a Dell PowerEdge 8450 server (CPU 700MHz, RAM 8G). ::: Optimization of the πvector is time consuming. The running time for reconstruction with or without optimizing πvector is 14,902 seconds and 213 seconds for SH2 alignment (44 sequences), respectively, on a Dell PowerEdge 8450 server (CPU 700MHz, RAM 8G) (Table [2](#T2){ref-type="table"}). In our program, the default πvector is calculated from the alignment while the user has the option to optimize the πvector for ancestral sequence reconstruction. Testing reconstruction ---------------------- Two different methods for simulations of the evolutionary process were used, as described in Methods, to test the reliability of the reconstruction results. In the first simulation method, starting from a randomly generated root sequence, we simulated the evolutionary process to obtain leaf node sequences based on a tree and a rate matrix. This process was repeated 100 times for a given root sequence Rto produce 100 alignments consisting of all leaf node sequences. For each of the 100 alignments, we used the marginal reconstruction method to obtain an amino acid probability vector for each site at the root. To reduce sampling variance, the amino acid probability vector was averaged over the 100 simulation trials. At each site, the amino acid with the highest average probability was chosen as our result of the \"reconstructed amino acid\" at that site. All \"reconstructed amino acids\" formed the reconstructed sequences R\'. There is no difference between Rand R\', that is, the accuracy of reconstruction is 100% for the tree shown in Figure [2](#F2){ref-type="fig"}. For each individual simulation and its reconstruction, we checked the amino acid with the highest probability in the reconstructed probability vector of the root. If it is indeed the \"reconstructed amino acid\", the prediction for that simulation is correct according to the average reconstructed results. The fraction of individual predictions that are correct according to the average reconstructed results is almost always higher than the average probability of the \"reconstructed amino acid\", suggesting that the average probability of the \"reconstructed amino acid\" gives a lower estimation of the reconstruction reliability (Figure [4a](#F4){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### *a) Correlation between the average probability of \"the reconstructed amino acid\" and the fraction of correct predictions. b) Correlation between the fraction of correct predictions and average α~AB~at each site.The protein family used here is the PDZ domain. Red filled points are sites with incorrect reconstruction. ::: ![](1471-2148-4-33-4) ::: For the second simulation method, we introduced rate heterogeneity across sites with structural and functional constraints \[[@B21]\]. For the same tree, the accuracy of reconstruction was about 90%. Sites with larger substitution rates are expected to have less reliable reconstructions. Figure [4b](#F4){ref-type="fig"} shows the relationship between the average α~AB~and the fraction of individual predictions that are correct according to the \"reconstructed amino acid\". Sites with incorrect \"reconstructed amino acids\" all have large α~AB~values. These values reflect the difficulty of reconstructing sites with large numbers of substitutions. The probabilities of the \"reconstructed amino acids\" are all small for sites with incorrect reconstructions (less than 0.15), suggesting that the information content of the reconstruction is low. The second simulation method was also used to test ANCESCON along with the reconstruction programs from PAML \[[@B30]\], PHYLIP \[[@B31]\] and PAUP\ \[[@B32]\]. All tree topologies used in reconstruction tests were inferred from real alignments. All original root sequences were taken from PDB database \[[@B33]\]. We had three different types of alignment testing sets. The first testing set used the same tree topology but different root sequences to generate 100 alignments (for details see Methods). The second testing set used the same root sequence but different tree topologies. The third testing set randomly selected a root sequence and a tree topology to generate 100 alignments. After 100 alignments were generated, we reconstructed the root sequence for each alignment and found the consensus root sequence for the 100 reconstructed root sequences. Finally, the consensus root sequence was compared with the original root sequence to calculate the reconstruction accuracy, i.e. the fraction of correctly reconstructed sites for the root sequence. In addition, for the third test, the paired t-test was used to calculate the one-tail probability between ANCESCON and other three methods. In order to make different tree topologies comparable, those trees were scaled to make the average distance from root to all leaf nodes (d~a~) the same for all trees and equal to the tree of pii1 (a signal transduction protein) (d~a~= 4.23). If d~a~was too small (e.g. 0.5), the reconstruction accuracy was always close to 1 for all reconstruction methods used. The value d~a~= 4.23 was large enough to generate diverse sequences to differentiate 4 different ancestral sequence reconstruction methods. For ANCESCON we had 3 different parameter settings, which included site-specific rate factors estimated by maximum likelihood method (α~ML~), Alignment-Based rate factors (α~AB~) and no rate factors (equal rates among sites). Different parameters were also used for the reconstruction programs from PAML and PHYLIP to find their best reconstructions. For PAML, reconstruction was tested with parameter α(rate factor) estimated from alignment and without α. For PHYLIP, 4 different parameter settings were tried, which were combinations of with/without αestimated from alignment by PAML and with/without branch length dwelling in input tree topology. For PAUP\, default settings were used. Table [3](#T3){ref-type="table"} shows a comparison of the reconstruction accuracy for these 4 methods. The reconstruction accuracy of ANCESCON with α~ML~is higher than the other three methods in almost every test. Also the reconstruction accuracy of ANCESCON with α~AB~and without αis comparable with PAML and PHYLIP methods and is much better than PAUP\. For the first testing set, the best average accuracy for ANCESCON is about 0.5, while the best average reconstruction accuracies for PAML, PHYLIP and PAUP\* are 0.45, 0.39 and 0.32 respectively. Testing set 2 and 3 produce similar results. Using the paired t-test in the third testing set, we show that ANCESCON method with α~ML~gives significantly better reconstruction than the other 3 methods. Because the site-specific α~ML~is very close to the true mutation rate at a site (Figure [1b](#F1){ref-type="fig"}), using the site-specific α~ML~can improve our ability to reconstruct the amino acids for ancestral sequences correctly. These reconstruction tests suggest that ANSCESCON may be a better tool to reconstruct ancestral sequences compared to PAML, PHYLIP and PAUP\* if the given alignment contains more diverse sequences. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Ancestral sequence reconstruction accuracy by different programs ::: Root Seq. Tree Leaf Node Num. Methods --------------------- ----------- ---------------- ---------- -------- ---------- -------- ---------- -------- -------- -------- ------ ------ 1em2 pii1 25 0.45 0.32 0.35 0.41 0.37 0.29 0.27 0.21 0.29 0.26 1g9o pii1 25 0.56 0.46 0.47 0.53 0.53 0.51 0.54 0.40 0.51 0.47 1rgg pii1 25 0.60 0.42 0.47 0.60 0.62 0.47 0.58 0.32 0.56 0.47 1sgt pii1 25 0.38 0.34 0.33 0.33 0.32 0.32 0.33 0.27 0.33 0.32 1zm2 pii1 25 0.33 0.29 0.3 0.28 0.25 0.21 0.25 0.21 0.27 0.16 2a8v pii1 25 0.62 0.45 0.42 0.56 0.55 0.44 0.46 0.28 0.50 0.36 2ctb pii1 25 0.53 0.40 0.39 0.41 0.38 0.24 0.24 0.21 0.29 0.22 Average accuracy 0.496 0.383 0.390 0.446 0.431 0.354 0.381 0.271 0.393 0.323 2ctb gef 27 0.54 0.37 0.38 0.35 0.35 0.29 0.17 0.24 0.22 0.22 2ctb LacI 54 0.66 0.64 0.57 0.44 0.37 0.49 0.35 0.42 0.33 0.34 2ctb pdz 39 0.54 0.41 0.42 0.44 0.39 0.22 0.34 0.18 0.32 0.22 2ctb ph 30 0.79 0.74 0.75 0.53 0.55 0.45 0.25 0.43 0.37 0.32 2ctb pii1 25 0.53 0.40 0.39 0.41 0.38 0.24 0.24 0.21 0.29 0.22 2ctb ptb 29 0.58 0.39 0.43 0.39 0.38 0.29 0.23 0.26 0.24 0.23 2ctb sh2 34 0.61 0.42 0.40 0.43 0.40 0.30 0.22 0.20 0.27 0.22 2ctb sh3 43 0.83 0.82 0.80 0.62 0.55 0.69 0.45 0.66 0.46 0.54 2ctb GST 140 0.76 0.73 0.73 @ @ \# \# 0.47 0.38 0.33 Average accuracy^&^ 0.635 0.524 0.518 0.451 0.421 0.371 0.281 0.325 0.313 0.289 1em2 pdz 39 0.45 0.35 0.36 0.44 0.44 0.29 0.43 0.23 0.4 0.24 1g9o pii1 25 0.56 0.46 0.47 0.53 0.53 0.51 0.54 0.40 0.51 0.47 1rgg sh2 34 0.64 0.48 0.46 0.61 0.61 0.56 0.59 0.34 0.6 0.41 1sgt gef 27 0.49 0.39 0.40 0.48 0.44 0.42 0.44 0.36 0.45 0.41 1zm2 ptb 29 0.66 0.47 0.48 0.57 0.57 0.53 0.51 0.32 0.52 0.41 2a8v ph 30 0.81 0.78 0.81 0.71 0.74 0.60 0.61 0.50 0.65 0.50 2ctb LacI 54 0.66 0.64 0.57 0.44 0.37 0.49 0.35 0.42 0.33 0.34 Average accuracy 0.610 0.510 0.507 0.540 0.529 0.486 0.496 0.367 0.494 0.397 Probability^Δ^ 0.0026 0.0023 0.0248 0.0328 0.0007 0.0168 0.0001 0.0143 0.0005 All root sequences are taken from PDB database and the names listed in the table are PDB IDs. Tree topologies for gef (guanine nucleotide exchange factor), LacI (PurR/LacI family of bacterial transcription factors), pdz, ph, pii1 (a signal transduction protein), ptb, sh2, sh3 and GST (glutathione S-transferase) are inferred from multiple sequence alignments chosen from Pfam database (version 7.3). All tree topologies are generated from real alignments and the distances are rescaled in order to make the trees comparable. The value in this table represents the accuracy of reconstruction, i.e. the fraction of correctly reconstructed sites for the root sequence. The best reconstruction accuracy in each test is shown in bold. α~ML~means that the site-specific rate factors were estimated by maximum likelihood method. α~AB~means that the site-specific rate factors were estimated by our empirical equation based on the given alignment (for details see Methods). -αmeans that the rate factors were not considered in reconstruction. +αmeans that the rate factors were considered in reconstruction. +L means that branch lengths of the input tree were used in reconstruction, while -L means that branch lengths were estimated by the reconstruction program itself. @: tree topology for GST had 140 leaf nodes that were too many for PAML to run through. ^\$^: rate factors estimated by PAML were used by PHYLIP in ancestral sequence reconstruction. \#: tree topology for GST had 140 leaf nodes, which were too many for PAML to estimate rate factors for GST. ^&^:GST is excluded in calculation of the average. ^Δ^: paired t-test method \[40\] was used to estimate the one-tail probability between ANCESCON and the other three reconstruction methods. ::: Ancestral sequences used in homology detection ---------------------------------------------- Thirty-eight OB (Oligonucleotide/oligosaccharide binding)-fold \[[@B34]\] proteins and ten other alignments (adenylyl kinase, gef, globin, pdz, ph, ptb, ras, sh2, sh3 and subtilase) from the Pfam database (version 7.3) \[[@B29]\] were chosen to perform homology detection tests. Given an alignment with Nsequences, we had four different methods, \"BEST\", \"SECOND BEST\", \"SHUFFLE\" and \"RANDOM\", to generate another N-1 sequences (for details see Methods). For each combined alignment (2N-1 sequences), PSI-BLAST \[[@B35]\] searches were performed starting from each sequence and seeded with the combined alignment (-B option in the program BLASTPGP, e-value cutoff 0.01), and all found hits were pooled together. The benchmark experiment was PSI-BLAST seeded with the native alignment (Nsequences). For each type of the four combined alignments, we checked hits not found by the native alignments. New hits were verified to be true positives or false positives by running PSI-BLAST or HMMER \[[@B36]\], followed by manual inspections. Using the 48 native alignments, a total of 13973 hits were found by the benchmark. Compared to the benchmark, the \"BEST\" method detected 120 new homologs and the other three methods found 69, 74 and 9 new homologs, respectively (Figure [5](#F5){ref-type="fig"}). Among those new homologs, \"BEST\", \"SECOND BEST\", \"SHUFFLE\" and \"RANDOM\" methods had 3, 2, 6 and 3 false positives, respectively (Figure [5](#F5){ref-type="fig"}). Also, \"BEST\", \"SECOND BEST\", \"SHUFFLE\" and \"RANDOM\" methods missed 61, 1070, 60 and 7811 homologs as compared to the benchmark. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Comparison of \"BEST\", \"SECOND BEST\", \"SHUFFLE\" and \"RANDOM\" methods in the number of new homologs detected when compared with the benchmark experiment.The methods are defined in \"Methods\" section. The blue portion of the bar shows the number of true positives. The red portion of the bar shows the number of the false positives. ::: ![](1471-2148-4-33-5) ::: Adding non-native sequences to the native alignment results in a change of sequence profile for PSI-BLAST searches. Random sequences can dilute the position-specific amino acid exchange characteristics of native alignments. This effect should not improve the profile. Indeed, few new homologs are found by the \"RANDOM\" method. However, sequences generated by shuffling each position of the native alignment have the same conservation properties as the native alignment, and the \"SHUFFLE\" method detects a total of 74 new homologs. Two effects may account for this finding. First, addition of shuffled sequences to the native alignment can slightly change the estimates of pseudocount frequencies of amino acids and thus the position specific scoring matrix \[[@B35]\]. Second, the new version of PSI-BLAST program uses composition-based statistics with e-value estimation related to the composition of the query sequence \[[@B37]\]. Each shuffled sequence has its own amino acid composition that is different from the native sequences. This difference can affect the e-values of hits. The \"BEST\" method detects the most number of new homologs, suggesting that the reconstructed ancestral sequences resemble the native sequences. Ancestral sequences may therefore be more similar to some remote homologs than to the native sequences. The \"SECOND BEST\" method detects less new homologs than the \"BEST\" method but more than the \"RANDOM\" method, suggesting that the second most probable amino acids in reconstruction can still reflect some properties of native sequences. Table [4](#T4){ref-type="table"} shows homology detection results of OB-fold structures using reconstructed ancestral sequences. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Homology detection results of OB-fold structures using reconstructed ancestral sequences ::: SCOP Superfamily/family PDB structure New homologs NCBI annotation ------------------------------------------------------------------------------------ ----------------- -------------- ---------------------------------------------------- Nucleic acid-binding proteins/ Anticodon-binding domain 1b7yB, 39--151 N/A \- 1b8aA, 1--102 N/A \- 1bbuA, 64--151 13431467 DNA polymerase II small subunit 15598836 DNA polymerase III, alpha chain 1c0aA, 1--106 11261591 DNA polymerase III, alpha chain 11499379 conserved hypothetical protein 1169392 DNA polymerase III alpha subunit 118794 DNA polymerase III alpha subunit 13620707 putative DNA polymerase III, alpha chain 14194684 DNA polymerase III alpha subunit 14194702 DNA polymerase III alpha subunit 14195653 DNA polymerase III alpha subunit 14195659 DNA polymerase III alpha subunit 15594924 DNA polymerase III, subunit alpha 15598836 DNA polymerase III, alpha chain 15601899 DnaE 15642243 DNA polymerase III, alpha subunit 15669005 M. jannaschiipredicted coding region MJ0818 15679404 DNA polymerase delta small subunit 3914611 ATP-dependent DNA helicase recG 1cuk, 1--64 N/A \- 1e1oA, 64--148 11261591 DNA polymerase III, alpha chain XF0204 14194684 DNA polymerase III alpha subunit 1fguA, 181--298 15219507 hypothetical protein 15230563 putative protein 15790309 Vng1255c from Halobacteriumsp. 6166145 DNA polymerase III alpha subunit 8778702 T1N15.20 1fl0A 10957481 hypothetical protein 1g51A, 1--104 14520587 hypothetical protein 14591565 hypothetical protein 15595886 hypothetical protein 3914638 ATP-dependent DNA helicase recG 1otcB, 36--126 N/A \- 1quqA, 62--152 15387767 probable replication protein a 28 Kd subunit 1qvcA, 1--114 N/A \- Nucleic acid-binding proteins/Cold shock DNA-binding domain like 1a62, 48--125 N/A \- 1ah9 N/A \- 1bkb, 75--139 15790688 translation initiation factor eIF-5A; Eif5a 1c9oA 6014735 Cold shock protein CspSt 1csp N/A \- 1d7qA N/A \- 1mjc N/A \- 1rl2 N/A \- 1sro 15671445 N utilization substance protein A 15794781 N utilisation substance protein A 15803711 transcription pausing; L factor 2eifA, 73--132 N/A \- Nucleic acid-binding proteins/DNA ligase, mRNA capping enzyme, domain2 1a0i, 241--349 N/A \- 1dgsA, 315--400 N/A \- 1ckmA, 238--302 N/A \- 1fviA, 190--293 N/A \- Nucleic acid-binding proteins/Phage ssDNA-binding proteins 1gpc N/A \- 1gvp N/A \- 1pfs N/A \- Nucleic acid-binding proteins/RNA polymerase subunit RBP8 1a1d N/A \- Staphylococcal nuclease/Staphylococcal nuclease 1eyd 13422779 aldose 1-epimerase \* Bacterial enterotoxins/Bacterial AB5 toxins, B units 1c4qA N/A \- 1prtF N/A \- Bacterial enterotoxins/Superantigen toxins 1an8, 19--94 N/A \- TIMP-like/Tissue inhibitor of metalloproteases 1ueaB, 14--106 N/A \- Inorganic pyrophosphatase/ Inorganic pyrophosphatase 2prd N/A \- MOP-like/BiMOP, duplicated molybdate-binding domain 1b9mA, 127--262 10639288 probable ATP-binding protein 10955070 AgtA 1175513 Putative ferric transport ATP-binding protein afuC 15598450 probable ATP-binding component of ABC transporter 3978166 ATPase FbpC 4895001 glucose ABC transporter ATPase \* Histidine kinase CheA, C-terminal domain/ Histidine kinase CheA, C-terminal domain 1b3qA, 540--671 N/A \- \* Putative false positives as assessed by manual inspection. ::: Prediction of functional sites ------------------------------ Ten well-studied protein families (adenylyl kinase, gef, globin, pdz, ph, ptb, ras, sh2, sh3 and subtilase) from the Pfam database (version 7.3) \[[@B29]\] were selected to test the prediction of functional sites. To define functional sites, we considered residues falling within 5Å of any ligand to be functionally important (i.e. AP5 for adenylyl kinase). As a simple quantification of prediction accuracy, we counted the number of predictions that lie within 5Å from the ligands and consider these sites to be true positives. Our method intends to identify those sites with high similarity within individual sub-trees and high variation among sub-trees. These sites are likely to contribute to functional specificity. Based on a tree partition and the reconstructions at the cutting nodes (details see Methods), we have developed a measure called specificity score (equation (27)). We expect that both highly variable sites and highly conserved sites tend to score low in our method. Ten top-ranking sites were selected as our predicted functional sites for each family. For comparison, we also implemented a simple conservation (SC) method \[[@B25]\], the evolutionary trace (ET) method \[[@B26]\] and the conservation difference (CD) method \[[@B21]\] on the 10 protein families. The results are shown in Table [5](#T5){ref-type="table"}. Here, the results from these three methods tend to include invariant or highly conserved sites while the result from our method scores those sites low. Still, the number of true positives of our method is comparable to other methods for several families. For some protein families, such as gef, pdz and subtilase, our method predicts no fewer functional residues than the other three methods. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Comparison of the true hits among the top 10 predicted sites for ANCESCON, evolutionary trace (ET), simple conservation (SC), and conservation difference (CD) methods ::: Protein Family PDB ID^\#^ Ligand/ substrate Number of sites \* \\ \\\* ANCESCON ET SC CD ---------------- ------------ ------------------- ----------------- ---- ------ -------- ---------- ----- ----- ---- adkinase 1aky AP5 188 42 20 18 3 9.5 9.1 8 gef 1bkd H-Ras 245 47 4 0 3 3 3 2 globin 1a6g HEM 147 21 1 1 2 5.5 6 6 pdz 1be9 \+ 81 15 2 1 6 4 4 2 ph 1mai I3P 109 11 2 0 2 2 3 2 ptb 1shc PTR 157 27 2 1 6 5 5 9 ras 821p GTN 185 29 10 9 2 5.6 8.7 5 sh2 1a09 ACE 83 17 2 1 3 5 4 4 sh3 1nlo ACE 57 9 1 1 2 5 4 0 subtilase 1av7 SBL 278 22 8 4 5 4.6 3.8 4 ^\#:^Representative protein structure \: Number of sites within 5Å to ligand or substrates \\: Number of invariant sites, which may contain gaps \\\: Number of invariant sites within 5 Å to ligand or substrates +: C-terminal peptide of protein CRIPT ::: Figure [6](#F6){ref-type="fig"} shows the mapping of our predictions on the structure for the PDZ domain family. In green color is the ligand and in red color are the functional residues predicted by our method. Six of the predicted residues are within 5Å to the peptide ligand. Nine of the predicted residues are around the ligand binding area. Only one is distant from the ligand (Figure [6](#F6){ref-type="fig"}). ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Mapping top 10 predictions by ANCESCON to PDZ domain (PDB ID: 1be9) \[50\].The color code scheme: ligand is shown in green and the predicted functional residues are shown in red. ::: ![](1471-2148-4-33-6) ::: Another example is the family of adenylyl kinases. Our method identified 3 residues within 5 Å to the ligand while the other 3 methods identified more such residues, most of which are in highly conserved positions such as the catalytic residues. Highly conserved residues, however, are not selected by our method since our measure is designed to emphasize on sites contributing to specificity. Figure [7](#F7){ref-type="fig"} shows the N-terminal part of the alignment of adenylyl kinases, with our predictions highlighted in red and orange colors. The evolutionary tree for the alignment is shown in Figure [8](#F8){ref-type="fig"}. The first cutting layer (for details see Methods) results in two well-separated sub-trees. Functional annotations suggest that they contain enzymes with different substrate preferences: adenylyl kinases and uridylate kinases, respectively. Three residues (27, 54 and 89) from our predictions (red colored in Figure [9](#F9){ref-type="fig"}) contribute to substrate-binding specificity, as have been noted before in the structural studies of the UMP kinases \[[@B38]\]. Figure [9](#F9){ref-type="fig"} highlights our predicted functional residues on the adenylyl kinase protein structure. Most of our predictions fall within the specificity pocket. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### A partial alignment of the N-terminal part of adenylyl kinases.Sites colored in red are our predictions that are within 5Å from the ligand. Sites colored in orange are our predictions more than 5Å apart from the ligand. ::: ![](1471-2148-4-33-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### The evolutionary tree for the adenylyl kinase family generated by \"Weighbor\".The first cutting layer is shown. Evolutionary distances are shown to scale. ::: ![](1471-2148-4-33-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Mapping top 10 predictions by ANCESCON to adenylyl kinase domain (PDB ID: 1aky) \[47\].The color code scheme: ligand is shown in green and the predicted functional residues are shown in red. ::: ![](1471-2148-4-33-9) ::: Conclusions =========== We developed a package (ANCESCON) to reconstruct ancestral protein sequences that takes into account the variation of substitution rates among sites. Two methods were proposed to estimate site-specific evolutionary rates (α), namely Alignment-Based rate factor (α~AB~) and rate factor αestimated by maximum likelihood (α~ML~). Consideration of rate variation among sites can alleviate the underestimation of evolutionary distances. Accuracy of ancestral sequence reconstruction by our method is higher than that of PAML, PHYLIP and PAUP\* when the given alignment contains more diverse sequences. We show that reconstructed ancestral sequences help to improve detection of distant homologs and prediction of functional sites with specificity. Methods ======= Transition probability and likelihood calculations -------------------------------------------------- For all models discussed in this paper, we assume all sites in an alignment evolve independently and according to a homogeneous, stationary and time reversible Markov process. The probability of an amino acid ito be replaced by amino acid jafter a time interval tis P~ij~(t). The transition probability matrix of 20 amino acids is written as P(t), which can be calculated as P(t) = exp(Qt) (2) Here, Qis the rate matrix. The non-diagonal elements q~ij~are the instantaneous rates of change from amino acid ito amino acid jand diagonal elements q~ii~are such that each matrix row sums up to 0. Qcan be calculated by: Q= S\* diag(π) (3) Sis the matrix of amino acid exchangeability parameters \[[@B39]\]. π~i~is the equilibrium frequency for amino acid i. Time reversibility implies that Sis a symmetric matrix. In our program, the Smatrix is taken from Whelan and Goldman \[[@B39]\] and the default πvector is estimated from the given alignment. Qcan be decomposed into eigenvalues (λ~i~) and eigenvectors (*u~i~). ![](1471-2148-4-33-i2.gif) U= (u~1~, \..., u~20~) (5) P~ij~(t) can be calculated using the following equation, ![](1471-2148-4-33-i3.gif) The likelihood function \[[@B8]\] for an evolutionary tree T shown in Figure [10](#F10){ref-type="fig"} is: ::: {#F10 .fig} Figure 10 ::: {.caption} ###### An evolutionary tree topology.Nodes C, D, E and F represent given protein sequences, while nodes A and B represent ancestral protein sequences, i.e. unknown sequences. d~YZ~represents the evolutionary distance between nodes Y and Z. ::: ![](1471-2148-4-33-10) ::: ![](1471-2148-4-33-i4.gif) Here, ![](1471-2148-4-33-i5.gif) is the equilibrium frequency of the amino acid at the node A. ![](1471-2148-4-33-i6.gif) is the transition probability from the amino acid at node A to the amino acid at node B after an evolutionary distance d~AB~. Considering that each site ihas a rate factor α~i~\[[@B13],[@B18]\], we have: ![](1471-2148-4-33-i7.gif) tin equation (6) can be expressed as: t= α·d* (9) dis the evolutionary distance and αis rate factor. The following restriction on the vector αholds: ![](1471-2148-4-33-i8.gif) Here, Kis the number of sites. Alignment-Based Rate Factor α(α~AB~) and Rate factor αestimated by Maximum Likelihood (α~ML~) --------------------------------------------------------------------------------------------------------- Our program supports two methods to estimate a rate factor for each site: Alignment-Based rate factor α(α~AB~) and Maximum Likelihood-estimated rate factor α(α~ML~). The estimation of α~AB~is empirical and based on the observation that the substitution rate at a site is correlated with the conservation of the site, which, in turn, is correlated with the average transition probability among the amino acids at that site. Conserved sites are dominated by highly similar amino acids and thus have high average transition probabilities among the amino acids. The algorithm to calculate α~AB~is as follows: 1\. Set tequal to 1.0 and use equation (6) to calculate a transition probability matrix Pfor 20 amino acids. Equation, ![](1471-2148-4-33-i9.gif), is used to compute a symmetric matrix P\'. 2\. Calculate the average transition probability for each site and take the reciprocal: ![](1471-2148-4-33-i10.gif), where ![](1471-2148-4-33-i11.gif) is the number of non-gapped amino acid pairs in site iand the denominator is the sum over the transition probabilities between all amino acid pairs (j,k) at a site i. 3\. For invariant sites, C~i~is set to 0 to make it consistent with the Maximum Likelihood estimation. 4\. Equation (11) is used to calculate α~AB~, so that equation (10) holds. ![](1471-2148-4-33-i12.gif) If an evolutionary tree is assumed for the alignment, we can estimate the α~ML~factors by maximizing the likelihood (equation (8)) for each site: ![](1471-2148-4-33-i13.gif) If some sites are highly variable, the α~ML~at those sites can be very large, as has been previously noticed \[[@B18]\]. We consider these rate factors to be outliers. For these sites, we have observed that likelihood changes very little over a wide range of the αvalues. An empirical method is used to reduce the values of α~ML~outliers, guided by the α~AB~values. a Z-score of the ratio of α~ML~to α~AB~is calculated for each site except invariant sites: ![](1471-2148-4-33-i14.gif) ![](1471-2148-4-33-i15.gif) Here, ![](1471-2148-4-33-i16.gif) is the ratio of ![](1471-2148-4-33-i17.gif) to ![](1471-2148-4-33-i18.gif) for site i; ![](1471-2148-4-33-i19.gif) is the number of sites excluding the invariant sites. If Z~i~is greater than 3, it is reduced to 3 by decreasing the value of ![](1471-2148-4-33-i17.gif). We repeat this procedure until no Z~i~for any site iis greater than 3. After removing the outliers, we scale the ![](1471-2148-4-33-i17.gif) values so that equation (10) holds. Amino acid frequency vector πoptimization ------------------------------------------- Two methods are implemented to estimate the equilibrium frequency vector π, one derived directly from the given alignment (Alignment-Based πor π~AB~) and the other estimated by Maximum Likelihood (π~ML~). The likelihood for the entire alignment is a function of πwith 19 variables. A continuous minimization method by simulated annealing \[[@B40]\] is used to optimize π, with the objective function being the logarithm likelihood of the alignment. The simulated annealing is computationally intensive and is the major reason for the long CPU time given in Table [2](#T2){ref-type="table"}. Distance matrix calculation and tree inference ---------------------------------------------- A Maximum Likelihood approach is used to estimate the evolutionary distances among sequences, either considering rate variation across sites or not. The logarithm likelihood for replacing one protein sequence (A) with another protein sequence (B) after an evolutionary distance dcan be written as: ![](1471-2148-4-33-i20.gif) Here, ![](1471-2148-4-33-i21.gif) is the equilibrium frequency for the amino acid at site jin sequence A. ![](1471-2148-4-33-i22.gif) is the transition probability from amino acid at site jin sequence A to amino acid at site jin sequence B after an evolutionary distance α~j~·d. α~j~is 1 if all sites are assumed to evolve at the same rate; otherwise the α~AB~at site jis used for α~j~. An estimate of the evolutionary distance between two sequences is obtained by maximizing the likelihood function of equation (15): ![](1471-2148-4-33-i23.gif) Equation (16) can be solved by the bisection root-finding method \[[@B40]\]. After the distance matrix is calculated, the \"Weighbor\" method, i.e. weighted neighbor joining, is used to infer an evolutionary tree \[[@B10]\]. Ancestral sequence reconstruction --------------------------------- Two methods are implemented to reconstruct ancestral sequences. One is a marginal reconstruction method \[[@B4]\], and the other is a joint reconstruction method \[[@B5]\]. Below are their brief descriptions. The marginal reconstruction method \[[@B4]\] -------------------------------------------- We calculate P(A~r~\\|{A~l~}T), which is the conditional probability of amino acid A~r~at the root, given leaf node amino acid set {A~l~} and a tree T. Since time reversibility is assumed, any internal node can serve as a root. Using Bayes\' theorem, we have: ![](1471-2148-4-33-i24.gif) Here, P(A~r~) is used here instead of P(A~r~\\|T) because the frequency of the root amino acid A~r~, i.e. π~r~, does not depend on tree T. P({A~l~}\\|A~r~T) is the conditional probability of the known amino acids at the leaf nodes, given Tand A~r~. P({A~l~}\\|T) does not depend on A~r~, so it is calculated as a normalization constant for P(A~r~\\|{A~l~},T) terms over all 20 possible values of A~r~to make the sum equal to 1. For Figure [10](#F10){ref-type="fig"}, P({A~l~}\\|A~r~T) can be expanded as: ![](1471-2148-4-33-i25.gif) Here, ![](1471-2148-4-33-i6.gif) is the transition probability from amino acid at node A to amino acid at node B after an evolutionary distance d~AB~. Equation (18) can be calculated using a recursive method suggested by Felsenstein \[[@B8]\]. If rate factors are used in the reconstruction of the root sequence, we have: ![](1471-2148-4-33-i26.gif) Here, α~i~could be either α~AB~or α~ML~at site i. P(A~C~,A~D~,A~E~,A~F~\\| A~A~,T)~i~is the conditional probability P(A~C~,A~D~,A~E~,A~F~\\| A~A~,T) at site i. The joint reconstruction method \[[@B5]\] ----------------------------------------- The objective of a joint reconstruction method is to find the combination of amino acids for an internal node set {A~i~} that maximize the conditional probability of this amino acid combination, given the leaf node amino acid set {A~l~} and a tree T, P({A~i~}\\|{A~l~},T). Using the Bayes\' theorem, we have: ![](1471-2148-4-33-i27.gif) Because P({A~l~}\\|T) is the same for all amino acid combination at internal node set {A~i~} this problem becomes finding the maximum of P({A~l~}\\|{A~i~},T) \*P({A~i~}). The details of a fast algorithm to solve equation (20) can be found in Pupko et al. \[[@B5]\]. We also incorporated site-specific rate factors in this algorithm, in a similar way as equation (19) Gaps ---- Due to difficulties with the probabilistic models of gaps, a simplified empirical approach is used to alleviate the problem. We assume that gaps are \"supersede\" letters. Gaps are considered for each site independently. If a leaf node has a gap instead of an amino acid at a site, this node will be deleted from the tree for this site. After dealing with leaves, we check all internal nodes for children. If an internal node has no children or only one child due to the leaf removal because of gaps, it will be removed from the tree and a gap will be assumed as its reconstructed state. Simulations of evolutionary process ----------------------------------- Two methods of simulating amino acid substitution process were used to test the reliability of reconstruction, rate factors and evolutionary distance estimation. The first simulation method was based on a homogeneous time reversible Markov model. The parameters from Whelan and Goldman \[[@B39]\] were chosen for our model, including the equilibrium frequency vector πand the Smatrix. Given the length of a branch from a parent node to one of its child nodes and the amino acid for the parent node, we simulated the substitution process to generate an amino acid for the child node based on the transition probabilities that were calculated using equation (6). For the arbitrarily selected tree shown in Figure [2](#F2){ref-type="fig"}, we first generated a random sequence of 100 amino acids as the root sequence based on the amino acid frequencies from Whelan and Goldman \[[@B39]\]. We then simulated the random substitution process to obtain all leaf node sequences. This simulation was repeated 100 times. The resulting 100 alignments were used to test the reliability of the reconstruction result. In this simulation, each site evolved independently according to the same tree topology and branch lengths, thus there was no rate heterogeneity across sites. The second simulation method, based on a Z-score model, introduced rate variation across sites by using structural and functional information for a specific protein family \[[@B21]\]. We selected three protein families for the Z-score simulations under structural and functional constraints: pdz domain (Protein DataBank (PDB) ID: 1g9o) \[[@B41]\], trypsin (PDB ID: 1sgt) \[[@B42]\] and carboxypeptidase A (PDB ID: 2ctb) \[[@B43]\]. Given a rooted tree, the native sequence with known structure was used as the root sequence. Simulations were made along the tree to generate sequences at any internal node or leaf node. If the evolutionary distance from a parent node to a child node was d, the child sequence was obtained after l\*daccepted substitutions starting from the parent sequence, where lis protein sequence length. Simulations of the substitution process were repeated 100 times. For each site, the number of accepted substitutions was recorded and averaged over 100 simulations. Rate factors (observed α), representing site mutability, were calculated from these average substitution numbers, such that the average of rate factors is 1 (equation (10)). 100 simulated alignments were used to test the rate factor estimators (α~AB~and α~ML~), distance calculation methods and ancestral sequence reconstruction. Homology detection ------------------ ### Testing dataset 38 OB (Oligonucleotide/oligosaccharide binding)-fold \[[@B34]\] proteins with known structures were selected for homology detection test. OB-fold has a 5-stranded β-barrel structure. In the SCOP (Structure Classification of Proteins) database (version 1.55) \[[@B44]\], there are 7 OB-fold superfamilies. The superfamily of nucleic acid binding proteins is the most populated. Diversity of many OB-fold homologs extends beyond detection by automatic PSI-BLAST searches. Multiple sequence alignments of native sequences were obtained from PSI-BLAST searches starting from the 38 OB-fold sequences with known structures. We also selected 10 alignments (adenylyl kinase, gef, globin, pdz, ph, ptb, ras, sh2, sh3 and subtilase) from the Pfam database (version 7.3) \[[@B29]\] for homology detection test. ### Four different methods For each alignment with Nsequences, ancestral sequences for the N-1 internal nodes were reconstructed. The idea is to test whether adding more sequences to a native alignment can help homology detection. Four types of combined alignments were generated, adding different sets of N-1 sequences to the native alignment. In the first case, the added sequence at each internal node consisted of amino acids with the largest probability at each position. In the second case, the added sequences were made up of amino acids with the second largest probability. In the third case, we shuffled the native alignment at each position while keeping the gap pattern as in the native alignment. After shuffling, we added N-1 sequences resulted from the shuffling to the native alignment. In the fourth case, N-1 random sequences were generated with the overall amino acid frequencies of the native alignment. These four methods are named \"BEST\", \"SECOND BEST\", \"SHUFFLE\" and \"RANDOM\", respectively. Prediction of functional sites ------------------------------ Our objective is to find sites that are well conserved within each sub-tree, but show high variability between different sub-trees. These sites are likely to contribute to functional specificity \[[@B26],[@B45],[@B46]\]. ### Sequence datasets Multiple sequence alignments of ten protein families were chosen from the Pfam database (version 7.3) \[[@B29]\]. These families are: adenylyl kinase (adkinase) (representing structure PDB ID: 1aky; its ligand or substrate: AP5) \[[@B47]\], guanine nucleotide exchange factor (gef) (1bkd; H-Ras) \[[@B48]\], globin (1a6g; HEM) \[[@B49]\], pdz domain (1be9; C-terminal peptide of protein CRIPT) \[[@B50]\], ph domain (1mai; I3P) \[[@B51]\], ptb domain (1shc; PTR) \[[@B52]\], ras (821p; GTN) \[[@B53]\], sh2 domain (1a09; ACE) \[[@B54]\], sh3 domain (1nlo; ACE) \[[@B55]\] and subtilase (1av7; SBL) \[[@B56]\]. Most of these alignments contain many sequences. We pruned and clustered the sequences in each alignment according to the length and diversity. Representative sequences were kept and used for tree inference and ancestral sequence reconstruction. This procedure was done in three steps: 1) removing fragments, 2) single-linkage clustering and 3) complete-linkage clustering, as described below. 1\. For each family, there is a template sequence with known structure. The sequences, which cover less than 75% of the non-gapped positions in the template sequence with amino acids, were considered to be fragments and discarded. 2\. A sequence identity matrix was calculated for the remaining sequences. A single linkage clustering was done to form sequence groups at sequence identity threshold 0.8. For each group, we chose the longest sequence as a representative, discarding other members. This step reduced redundancy in the dataset. 3\. An average sequence identity was calculated for the remaining sequences. We used this average identity as a threshold for complete linkage clustering to form new sequence groups. Four groups with the largest sequence numbers were chosen to form our new alignment. Any group with the same number of sequences as the fourth group was also included in the new alignment. The purpose of this step is to keep the major sequence subgroups of a family while leaving out highly divergent sequences that might be deleterious for tree inference. ### Rooting The \"Weighbor\" method gives an unrooted tree. For our purpose of predicting functional sites, we need to find a point on the tree that serves as the root. We used a least-squares modification of the midpoint rooting procedure to define the root \[[@B57]\]. ### Tree partitioning The tree was partitioned into sub-trees at several levels and compared the amino acid usages within each sub-tree and among the sub-trees. For this partitioning, we \"cut\" the tree into a fixed number of equal-distanced layers, using the midpoint as the root (Figure [11](#F11){ref-type="fig"}). Several criteria were tried for selecting the distance between adjacent layers. Empirically we found that a simple partition of the tree into 5 layers usually gave the best results. If the average distance from the root to all leaf nodes is d~r~, then the distance between adjacent layers is d~r~/5 (Figure [11](#F11){ref-type="fig"}). Each place of a \"cut\" between the layers corresponds to a certain ancestral sequence. We term the location of a \"cut\" as a \"cutting\" node. The marginal reconstruction method was used to reconstruct amino acid probability vectors for all the cutting nodes (Figure [11](#F11){ref-type="fig"}). The reconstructed probability vector of a cutting node reflects the amino acid usages of the sub-tree under it. ::: {#F11 .fig} Figure 11 ::: {.caption} ###### An example showing the different cutting layers in a rooted tree.d~r~is the average distance from the root to all leaf nodes. Nodes iand jare neighboring cutting nodes. ::: ![](1471-2148-4-33-11) ::: ### Calculating specificity score for each site We use {L~K~} to represent the set of cutting nodes for layer L~K~, K= 0,1,5. {L~0~} is the root and L~1~is the closest layer to the root, etc. A dissimilarity score between any neighboring cutting node pair is calculated. The definition of a neighboring cutting node pair (i, j) (Figure [11](#F11){ref-type="fig"}) is: 1\. i∈ {L~K~} 2\. j∈ {L~K+1~} 3\. Node iis an ancestor of node j(all points on the path from jto root node are ancestors of node j), so that the distance between iand jis exactly d~r~/5. Each cutting node has only one ancestral cutting node neighbor. The dissimilarity score for cutting node jand its ancestral cutting node neighbor i,i.e. anc(j), at site mis defined as: ![](1471-2148-4-33-i28.gif) ![](1471-2148-4-33-i29.gif) and ![](1471-2148-4-33-i30.gif) are the reconstructed probabilities of amino acid Aat cutting node jand its ancestral cutting node neighbor i(anc(j)), respectively. Let ![](1471-2148-4-33-i31.gif), K= 1,\...,5 (22) Here, ![](1471-2148-4-33-i32.gif) is the average dissimilarity score for layer K. N~K~is the number of cutting nodes in layer K. The specificity score is defined as: ![](1471-2148-4-33-i33.gif) ![](1471-2148-4-33-i34.gif) reflects the difference of amino acid compositions among the major sub-trees defined by the first layer. ![](1471-2148-4-33-i35.gif) to ![](1471-2148-4-33-i36.gif) reflect the average difference of amino acid compositions within each sub-tree. If the amino acids are highly conserved within each sub-tree but show variability among the sub-trees, ![](1471-2148-4-33-i35.gif) to ![](1471-2148-4-33-i36.gif) are small and ![](1471-2148-4-33-i34.gif) is large, leading to a large value of S~m~. We set S~m~to 0 for invariant sites. We sort the sites by their specificity scores and choose the 10 top scoring sites as our predicted functional sites. Those predicted functional sites that lie within 5 Å from the ligand(s) are considered to be true positives. ### Comparison with other methods We compared our method with three other methods for prediction of functional sites. The first method (Simple Conservation or SC) is based on sequence conservation. Highly conserved sites are considered to be functional. For each family, we sorted the sites by positional conservation \[[@B25]\] and chose the 10 top-ranking sites as the predictions. There might be ties for sites. For example, if there were 5 sites tied at the tenth conservation value and only one of them was within 5Å from the ligand(s), then its contribution to the total number of \"correct predictions\" was 1/5. The second method is the evolutionary trace (ET) method \[[@B26]\], which partitions a sequence identity dendrogram into sub-trees at varying sequence identity thresholds. Sites that are invariant within each individual sub-tree are picked as functional sites. A higher identity threshold gives rise to more sub-trees and, since conserved sites are more frequent in the sub-trees with smaller sizes, lead to more predicted sites. ET analysis was performed from a low identity threshold to higher thresholds until the number of predicted sites was 10 or just above 10 (in the cases of ties). Ties were resolved similarly to the simple conservation method. The third method (conservation difference or CD) is based on the conservation differences between a native alignment and an alignment derived from the Z-score sequence design \[[@B21]\]. The basic idea was to differentiate sites conserved due to structural stability and sites conserved due to function. Since the pairwise potential in the Z-score design tends to weaken the conservation caused by function, functionally conserved sites tend to have a large conservation difference between the native alignment and the alignment of designed sequences. We chose 10 top ranking sites sorted by conservation difference as predictions by CD. Authors\' contributions ======================= NVG conceived and initiated the study. All authors took part in developing methods and designing experiments. WC wrote the source code and JP analyzed the data. All authors read and approved the final manuscripts. Acknowledgements ================ We thank Lisa Kinch, James Wrabl and Hua Cheng for their useful comments. This work was supported by the NIH grant GM67165 to NVG.	PubMed Central	2024-06-05T03:55:48.416571	2004-9-17	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522809/", "journal": "BMC Evol Biol. 2004 Sep 17; 4:33", "authors": [ { "first": "Wei", "last": "Cai" }, { "first": "Jimin", "last": "Pei" }, { "first": "Nick V", "last": "Grishin" } ] }
PMC522810	Background ========== The advent of DNA microarrays has provided the science community with a tool to concurrently examine the expression of thousands of genes within a given cell or tissue type, thus providing a platform for future diagnoses and prognostic analyses of disease with gene-level specificity \[[@B1],[@B2]\]. Microarray technology is progressing rapidly as better sequencing and prediction algorithms allows for more refined gene prediction. This has prompted the evolution of the probe sets contained within the array chips over the past few years, in both oligonucleotide and cDNA arrays \[[@B3]\]. This microarray platform expansion has hindered the direct comparison between numerous datasets of a given phenotype that have been produced using several generations of arrays. In microarray analyses of disease, having a large number of samples better accounts for the biological variability between individuals and therefore increases the power to enhance and define a pathogenetic model for that disease. Due to the considerable expense of microarray chips and the equipment required, along with the common problem of sufficient sample acquisition, being able to combine and compare datasets from various laboratories and across all microarray generations would be a benefit to the entire biomedical community. The constant evolution of microarrays has thus resulted in a significant hindrance to their power as a research or diagnostic tool by dividing datasets according to platform and seemingly limiting their interarray comparability. With the large number of microarray datasets available in the public domain for distinct disease phenotypes from various microarray platforms, cross-platform comparisons can currently be attempted. For example, a number of laboratories have been studying diagnostic pediatric acute lymphoblastic leukemia (ALL) samples from human bone marrow on both oligonucleotide and cDNA microarrays and depositing the raw intensity values into the public domain. Additionally, in 2002 at St. Jude Children\'s Research Hospital, Yeoh et al. generated a list of genes that have distinct expression levels for various karyotypic and phenotypic aberrations common to pediatric ALL \[[@B4]\]. This set of genes has been useful as a prognostic profile for ALL by identifying subclasses of the cancer using microarray technology \[[@B4]\]. Importantly, the St. Jude\'s gene list has been validated on independent datasets both within their own lab and in an independent laboratory by Kohlmann et al. in 2004. They used the Yeoh et algene list to successfully segregate the various subclasses of adult ALL \[[@B5]\]. Consequently, this disease provides an excellent model for testing interarray comparability using one common gene list. ALL is the most common pediatric malignancy comprising over 75% of the annual diagnoses of leukemias in children \[[@B6]\]. In the United States, the outcome for children with ALL has improved dramatically over the past thirty years with the long-term disease-free survival (LTDFS) rates increasing from less than 10% in 1970 to over 80% today \[[@B7]\]. However, ALL still carries the risk of relapse in over 20% of patients \[[@B8]\]. ALL survival is largely due to a greater understanding of the risk factors that affect outcome, which has allowed for more intensity-tailored treatment following an assessment of the patient\'s risk \[[@B7]\]. Accurate segregation of patients into their proper risk group is critical to allow for a risk-stratified treatment that is effective enough to clear the disease and decrease the risk of relapse while minimizing the negative long-term side effects \[[@B7]\]. Factors that affect prognosis are age, sex, race, white blood cell count at diagnosis, phenotypic differences, such as T-cell versus B-cell lineage ALL and karyotypic alterations, such as the hyperdiploid karyotype and the translocations t(12;21)\[TEL-AML1\], t(4;11)\[MLL-AF4\], t(9;22)\[BCR-ABL\], and t(1;19)\[E2A-PBX\] \[[@B7],[@B9]\]. These genetic lesions can affect the individual\'s response to treatment. For example, patients with a hyperdiploid karyotype and those with the TEL-AML1fusion gene have a better prognosis than patients in the other subclasses \[[@B10]\]. The initial diagnosis and classification of ALL is currently revealed through multiple time-consuming and expensive tests often involving multiple laboratories \[[@B11]\]. Thus, a tool that could consolidate these tests into one diagnostic platform would be beneficial to both researchers and clinicians working with ALL. In this study, we sought to determine if datasets from different microarray platforms could be compared in a useful manner. We chose to study pediatric ALL because there is already a substantial pool of datasets freely available in the public domain. First, we collected pediatric ALL array data and cDNA array data generated in experiments from various laboratories. In addition to the collection of public data, we supplemented the hyperdiploid karyotype and the T-cell lineage ALL subclasses by expression profiling additional diagnostic pediatric ALL samples from a tumor bank in Children\'s National Medical Center in Washington, D.C. We used these independent datasets, including the one generated in our lab, to validate the gene predictors, as defined by Yeoh et al. (2002), for each of the aforementioned prognostic genetic variables. Cross-platform array validation was used successfully to ascertain the accuracy, sensitivity and specificity of the gene predictors for the prognostic variables. In addition, we have demonstrated the ability to compare datasets from different microarray platforms. To our knowledge, this is among the first known successful applications of this technique, along with the validation of the Yeoh et alpediatric ALL gene lists on adult ALL by Kohlmann, et al(2004) \[[@B5]\]. Interarray comparisons such as these will further enhance the ability to integrate data from several generations of microarray experiments and will help to break down barriers to the assimilation of existing datasets into a comprehensive data pool. Results and discussion ====================== Expression profiling of ALL diagnostic bone marrow -------------------------------------------------- To supplement the ALL subclasses that are under-represented in expression profiling thus far, we collected and extracted the total RNA from sixteen diagnostic bone marrow samples housed at Children\'s National Medical Center; seven of the hyperdiploid karyotype and nine with T-cell lineage. The extracts were then hybridized to Affymetrix U133A arrays and expression profiled as an independent training data set. Validation of the gene predictors using independent datasets spanning various array platforms --------------------------------------------------------------------------------------------- In order to validate the portability of gene predictors across microarray platforms we compared the accuracy with which the six prevalent ALL subclasses can be distinguished on disparate array platforms. To do this we used the discriminating gene lists (\~40 genes), which were provided by the comprehensive training ALL sample set analyzed and published by Yeoh et al. in 2002 \[[@B4]\]. In their study they hybridized RNA from ALL bone marrow samples to Affymetrix U95Av2 arrays. The resulting expression data were analyzed by multiple statistical methods to facilitate the generation of lists of genes that represent the greatest difference in expression between the ALL subclasses \[[@B4]\]. Yeoh et al. used both a training and test dataset in their analysis to first uncover the subclass-specific gene expression profiles and then to test their predictability on independent samples \[[@B4]\]. The genes are listed hierarchically, along with supplemental information about the statistical methods used, at <http://www.stjuderesearch.org/data/ALL1>. We then accumulated the ALL diagnostic bone marrow array data available in the public domain (Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Training and test datasets used to validate ALL subclass predictors ::: Training Sample Set Microarray Platform ----------------------------------------------------------------------------- --------------------------------- -------------------------------------- Yeoh et al. Cancer Cell, 2002, 1:133--143 Affymetrix HG\_U95Av2 Validation Sample Set Microarray Platform Predictors Armstrong et al. Nat. Genet., 2002, 30(1):41--7 Affymetrix HG\_U95Av2 Hyperdiploid, MLL-AF4, TEL-AML1 Mitchell et al. Unpublished data 2003 Affymetrix HG\_U133A Hyperdiploid, T-ALL Stephan DA, Golub TR. Unpublished data 2000 Affymetrix HuGene FL TEL-AML1, E2A-PBX1 Golub et al. Science, 1999, 286:531--7 Affymetrix HuGene FL T-ALL Ramaswamy et al. Proc. Natl. Acad. Sci. USA, 1999, 98(26):15149--54 Affymetrix Hu6800 and Hu35KsubA T-ALL Moos et al. Clin. Cancer Res., 2002, 8:3118--3130 cDNA TEL-AML1, MLL-AF4, BCR-ABL1, T-ALL Catchpoole et al. Unpublished data 2002. cDNA T-ALL ::: The independent datasets that we accumulated, including the one generated in our lab, spanned four different microarray platforms: Affymetrix HuGene FL, U95Av2, U133A and custom cDNA microarray platforms (Table [1](#T1){ref-type="table"}). To modify these test datasets into data that could be directly applied to the predictor gene lists from the U95Av2 arrays, we correlated the probe numbers between these different arrays and the U95Av2 set using the probe match spreadsheet, NetAffx, available at <http://www.affymetrix.com>. We then used the discriminating gene list for each subclass to extract the appropriate probes and their intensity values from the expansive expression data for each sample of the validation datasets independently. The level of similarity between the probe sets of the two different array platforms was evident through the number of genes within the 40 discriminators that could be found within the validation data (Table [2](#T2){ref-type="table"}). For example, the data published by Armstrong et al. (2002) was generated on the U95Av2 array platform \[[@B6]\]. Therefore, expression data for all 40 predictor genes could be correlated and represented in their corresponding MLL-AF4, TEL-AML1, and hyperdiploid datasets. Similarly, the U133A arrays that were used to generate expression data in our lab for the hyperdiploid karyotype and the T-cell lineage ALL contained probes representing the majority of the 40 discriminators, with 38 and 35 genes, respectively. The HuGene FL arrays contain significantly fewer probe sets in common with the selected predictors (from the later-generation U95Av2 microarrays). Accordingly, of the 40 original predictor probes, only 25 were present in the TEL-AML1dataset, 26 in the E2A-PBX1dataset and 13 in the T-cell dataset. The difficult task of matching probes from the Affymetrix gene chips with cDNA arrays was illustrative of the disparities between the probe sets within these two platforms. For example, the five cDNA predictors built using the datasets produced by Moos et al. (2002) and Catchpoole et al. (unpublished data) contained data for only ten genes or less from the predictor set gene list. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Prediction accuracies for ALL subclasses as determined by the different microarray platforms. ::: ALL Subclass Microarray Platform \# of Samples in the Dataset \# of Samples Representing the Predictor Subclass \# of Genes in Predictor (out of 40)^1^ Accuracy (%)^2^ Sensitivity (%)^3^ Specificity (%)^4^ ------------------ ------------------------- ---------------------------------- ------------------------------------------------------- --------------------------------------------- --------------------- ------------------------ ------------------------ Hyperdiploid Affymetrix U95Av2 43^a^ 5 40 97 80 100 Hyperdiploid Affymetrix U133A 16^b^ 7 38 94 86 100 T-ALL Affymetrix U133A 16^b^ 9 35 100 100 100 T-ALL Affymetrix HuGene FL 41^c^ 8 13 100 100 100 T-ALL Affymetrix Hu6800 20^d^ 10 30 95 100 90 T-ALL cDNA 52^e^ 7 5 98 86 100 T-ALL cDNA 9^f^ 3 29 100 100 100 TEL-AML1 Affymetrix U95Av2 43^a^ 9 40 91 67 97 TEL-AML1 Affymetrix HuGene FL 23^g^ 14 30 86 79 100 *TEL-AML1* cDNA 52^e^ 12 10 87 83 88 MLL-AF4 Affymetrix U95Av2 43^a^ 20 40 100 100 100 *MLL-AF4* cDNA 52^e^ 2 7 98 50 100 *E2A-PBX1* Affymetrix HuGene FL 23^c^ 2 26 96 50 100 ^1^With a few exceptions, the majority of the gene lists published by Yeoh et al(2002) contain 40 genes. ^2^The ability of the predictor to correctly classify the blinded test set into the correct subgroup ^3^(\# of positive samples predicted correctly)/(total \#of true positives) ^4^(\# of negative samples predicted correctly)/(total \#of true negatives) ^a^Armstrong et al. (2002) Nat. Genet. 30(1), 41--7. ^b^Mitchell et al. (2003) Unpublished data. ^c^Golub et al. (1999) Science286, 531--7. ^d^Ramaswamy et al. (2001) PNAS98(26), 15149--54. ^e^Moos et al. (2002) Clin Cancer Res. 8, 3118--3130. ^f^Catchpoole et al. Unpublished data. ^g^Stephan et al. (2000) Unpublished data. ::: To validate the gene predictors from Yeoh et al. (2002), using the aforementioned independent test datasets from various array platforms, we employed supervised learning methods using GeneCluster2 software. Prior to analysis, we formatted the discriminating gene expression values from the test datasets onto spreadsheets according to software instruction, and subsequently applied the data to the software. Genecluster2 then generated blinded predictions on the ALL samples of the test datasets through weighted voting with a leave-one-out methodology. This is accomplished by randomly removing one sample at a time from the test dataset of ALL samples and \"training\" a predictor gene profile to recognize similarities or disparities between the two classes based on the expression profiles of the samples for the genes of interest \[[@B11]\]. In this manner each sample is assigned to one of the two classes based on their expression pattern of the predictor genes. The prediction accuracy, sensitivity and specificity were calculated for each of the predictors from each array platform and are displayed in both figure [1](#F1){ref-type="fig"} and table [2](#T2){ref-type="table"}. The accuracy of our predictors ranged from 86%--100%, with a mean accuracy of 95%. The mean specificity of the predictors was 98%, ten of which provided a specificity of 100%. The sensitivity ranged from 50%--100%. The mean sensitivity was 83% (fig. [1](#F1){ref-type="fig"}, table [2](#T2){ref-type="table"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Summary of results of the various ALL subclass predictors tested. The predictors are organized according to microarray platform and the results are listed under each class in terms of the accuracy, sensitivity and specificity of the classification. ::: ![](1471-2164-5-71-1) ::: We saw a high accuracy from the predictors employing data from both U95Av2 and U133A arrays, attesting to the fact that nearly all of the 40 discriminating genes were present in the datasets, thus maximizing the possible prediction strength. In the case of the E2A-PBX1predictor (96%) and the MLL-AF4predictor (98%), the sensitivities were only 50%. In both cases there were only 2 samples out of the sample pool expressing the respective translocation and in both analyses one of the two was continuously classified incorrectly. This could be due to many factors, misdiagnosis or mislabelling of the sample, poor sample quality or differences in sample handling. It is difficult to draw a conclusion due to the fact that the samples were collected and processed in a laboratory outside of our own. Another problem with these two predictors may simply be the low sample number. Two samples may not provide enough strength to the classification by Genecluster2 simply due to the inability of such a low sample number to account for the biological variability that exists between patients that is independent of their subclass of ALL. The most surprising result was the high accuracy with which the gene lists could classify T-ALL, TEL-AML1and MLL-AF4from cDNA data considering the disparity between the probe sets of cDNA arrays and oligo arrays. The accuracies of the classifiers were: T-ALL (Catchpoole data), 100%; T-ALL (Moos data), 98%; MLL-AF4, 98%; and TEL-AML1, 87%. Therefore, it appears that the number of genes in the predictor gene list is much less of a factor in the predictor\'s classification accuracy than the number of samples representing the phenotype of interest, which supports the argument that being able to do cross-platform analyses to increase sample size is crucial for sensitive and specific class prediction using expression data. This is strongly illustrated by the cDNA predictors, which have few probes in common with the arrays used to generate the predictor gene list, but still classify the ALL samples with high sensitivity, specificity and accuracy. On the other hand the E2A-PBX1predictor and the MLL-AF4predictor had low sensitivities correlating to a low sample number in these groups. The high number of probes in common between the arrays used to generate these independent datasets and the arrays used to generate the predictor gene list, 26/40 and 40/40, respectively, were not able to rescue the low sensitivity of the predictor. Conclusions =========== Currently the vast majority of expression data from numerous labs is not being used to its highest potential as independent labs continue to move to more expansive array platforms rendering older datasets less informative in the context of new data. Increasingly, progressive technologies in genome databasing and chip construction are prompting this inevitable evolution of microarrays. Until data can be analyzed and directly compared across array platforms, the size of the data pools will remain small and isolated according to platform \[[@B12]\]. Here we have shown that the previously validated predictor gene list from Yeoh et al. (2002) withstands validation by testing the predictors using a leave-one-out strategy on all publicly available datasets as well as a dataset generated in our own lab regardless of the array platform used. This meta-data analysis of over 200 arrays from diagnostic ALL samples with hyperdiploidy, T-cell lineage and translocation status (previously confirmed through gold standard techniques), shows that expression profiling as an integrated platform is robust and that ALL data, and presumably other disease models, can be interplatform comparable. By validating the comparability between data from distinct microarray platforms we have demonstrated a tool that can enhance the statistical power provided by large sample sets. Thus, we can potentially develop and validate sensitive diagnostic tools based on large training sample sets, to allow for the rapid assignment of individualized therapy to improve disease outcome in pediatric ALL and other diseases. Methods ======= RNA extraction from bone marrow samples --------------------------------------- ALL diagnostic bone marrow samples were housed in a tumor bank in Children\'s National Medical Center in Washington, D.C. Mononuclear cells from diagnostic bone marrow aspirates were separated using density centrifugation on Cappel Lymphocyte Separation Medium (ICN Biomedicals, Aurora, Ohio) and immediately flash frozen according to manufacturer\'s instruction. A total of sixteen samples were obtained with IRB approval; seven with a hyperdiploid karyotype and nine samples of a T-cell lineage as confirmed by immunophenotyping. The frozen samples were placed directly in TRIzol reagent for RNAse-free thawing for total RNA extraction. We extracted a 10 μg--20 μg pellet of total RNA from each sample by centrifugation following phenol-chloroform extraction. The integrity of the resultant total RNA from each sample was quantified by gel electrophoresis before it was considered to be of good quality for cDNA synthesis. Samples were re-extracted if ribosomal bands were not visible. Expression profiling and support vector machine meta-analysis ------------------------------------------------------------- 10 μg of the extracted RNA from each sample was labelled and hybridized to an Affymetrix U133A array (Affymetrix, Santa Clara, CA) according to protocol as previously described \[[@B13]\]. Intensity values were calculated using Microarray Suite 5.0 (MAS 5.0) and expression values were adjusted to fall within the lower and upper limits of 1 and 45000 as described by Yeoh et al. (2002) \[[@B4]\]. To create a predictor that allows for the direct comparison between different generation Affymetrix arrays and cDNA arrays, we used the predictor gene list for each subclass provided by Yeoh et al. (2002) from Affymetrix U95Av2 microarrays. The 40 genes that showed the greatest mean difference in expression between the subclass of interest and the remaining subclasses was used as our predictor gene set. The gene lists and additional information, including the statistical metrics used to generate the gene list from the training set, can be viewed at: <http://www.stjuderesearch.org/data/ALL1>. To identify comparable data points between the gene lists from the training set (produced on the U95Av2 Affymetrix chip), and the expression values of samples provided by other public datasets on different generation Affymetrix arrays, we used the probe match function within NetAffx <http://www.affymetrix.com>. Data for these probe pairs in the validation sets were extracted and expression values were linearly adjusted to fall within 1--45000 \[[@B4]\]. Affymetrix probes were identified within cDNA data by a combination of BLAST sequence comparison <http://www.ncbi.nlm.nih.gov/BLAST/> and GenBank accession number queries. Ratios were log-transformed prior to analysis. GeneCluster2 (<http://www.genome.wi.mit.edu>; Center for Genome Research, MIT, Cambridge, MA) was used to perform blinded predictions on the validation dataset using weighted voting with a leave-one-out methodology. Accuracy, specificity and sensitivity values were then generated for each predictor, as a measure of the predictor\'s ability to correctly group the samples into their respective class in the validation sets. Authors\' contributions ======================= SAM carried out the accumulation of datasets, preparation of the predictors, data analysis and drafted the manuscript. SAM with the help of KMB and MMH participated in the sample preparation and hybridization to arrays of the in-house gene expression data. KMB also trained SAM and participated in the preparation of the predictors with SAM and also provided essential mentor support. MM participated in finding and extracting the public data and aided in drafting the manuscript. MMH was a tremendous help in editing the final manuscript. BL provided statistical guidance throughout the project. DC provided us with cDNA array data and aided in the analysis of the cDNA expression data. DC also provided much appreciated critical input throughout the entire project. DAS initiated and crafted the idea and provided the necessary mentorship along the way. Acknowledgements ================ This research was supported by NIH Grant 7R21CA095618. I would also like to give special thanks to Dachuan Guo and Belinda Cutri from Daniel Catchpoole\'s lab in Westmead, Australia for their work in preparing the cDNA arrays and handling the data. Dr. Catchpoole\'s lab was a tremendous help and support throughout this project and the write-up.	PubMed Central	2024-06-05T03:55:48.423973	2004-9-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522810/", "journal": "BMC Genomics. 2004 Sep 23; 5:71", "authors": [ { "first": "Stephanie A", "last": "Mitchell" }, { "first": "Kevin M", "last": "Brown" }, { "first": "Michael M", "last": "Henry" }, { "first": "Michelle", "last": "Mintz" }, { "first": "Daniel", "last": "Catchpoole" }, { "first": "Bonnie", "last": "LaFleur" }, { "first": "Dietrich A", "last": "Stephan" } ] }
PMC522811	Background ========== Candidaspecies are now among the most important pathogens especially for the immunocompromised host. They are the fourth most common organisms recovered from blood cultures in hospitalized patients \[[@B1]\]. Candida albicansthe most frequently isolated of the species, is a polymorphic organism. It can switch from a yeast form (blastospore) to a filamentous phase (hypha and pseudohypha) in response to a variety of external stimuli, including mammalian serum. Mutants defective in this serum response, also show a reduced capacity to cause disease in a murine model, \[[@B2]\] suggesting a virulence role for it. It is widely accepted that the production of the RNA components of ribosomes in eukaryotes proceeds through the transcription of large pre-RNA molecules by RNA polymerase I (Pol I), that get processed into the final large and small subsegments \[[@B3]\]. We have recently reported the unexpected finding that Candida albicanspolyadenylates some of its 25S ribosomal RNA (rRNA) and the polyadenylation site corresponded to the large subsegment 3\'-end maturation \[[@B4]\]. We also found that the concentration of the polyadenylated form of 25S was increased transiently by serum just prior to the appearance of filamentous forms, raising the possibility for a role in hyphal transformation. A question raised by these data was whether this event is unique for the large subunit rRNA or it represents a wider function. For example, is a DNA sequence upstream from the 25S subunit functioning serendipitously as a promoter for RNA polymerase II (Pol II), allowing transcription by this enzyme complex and subsequent polyadenylation. Such a process would not likely be important for ribosomal function. Similar involvement of other subunits on the other hand, would increase the likelihood that polyadenylation of rRNA plays a wider role in the biology of this yeast. We now report our observations related to the 18S subunit of rRNA, that indeed there are other polyadenylation sites located near the 3\'end of the 18S subunit. Similar to 25S, we found that a polyadenylated 18S transcript, similar in size to a processed mature molecule, is also enhanced early and transiently by serum, further strengthening the possibility of a regulatory role for polyadenylation in the germination process. Results ======= Cloning of poly A -extended 18S subsegments ------------------------------------------- We have found polyadenylation to occur both in yeast grown in YPD and in those exposed to serum for 5 minutes. In all cloned PCR products the number of adenines in the chain exceeded those in the poly-T primer (DT12) used to generate them. Furthermore, the anchor sequence assures us that we did not extend an inappropriately annealed primer. Unlike our data with the 25S subunit, where the attachment was to one of two thymidines one base apart at the reported maturation site \[[@B4]\], the polyadenylation sites near the 3\' end of 18S subunit were congregating either upstream or downstream of the reported maturation site \[[@B5]\] (Figure [1A](#F1){ref-type="fig"}) but none were at the reported site. Six of seven YPD exposed yeast polyadenylation sites were between positions 1625 and 1643, located 148 bases upstream from the reported 3\' end, whereas 3 of 4 polyadenylation sites downstream from the maturation site were from yeast exposed to serum. We were able to amplify a full-length clone from YPD exposed yeast and its site of polyadenylation was at position 1643 (Figure [1B](#F1){ref-type="fig"}) near the other polyadenylation sites of yeast exposed to YPD. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Polyadenylation sites of 18S rRNA subunit and a full length polyadenylated 18S clone.Polyadenylation positions on 18S subunit. (A) represents the clones derived from amplification with primers designed for the 3\'end of 18S. Underlined letters represent polyadenylation sites. Y or S over them indicate whether RNA came from yeast exposed to YPD (Y) or serum (S) and superscripted digits over them indicate the number of clones found at that position. The bold enlarged letters represent the sequence with the most polyadenylation sites. The \#1624 represents the position from 5\' end of 18S. (B) represents partial sequences at the 5\' and 3\' ends of a full length clone of a polyadenylated 18S molecule. ::: ![](1471-2199-5-17-1) ::: Serum enhancement of polyadenylation ------------------------------------ Similar to 25S, polyadenylation of 18S subunit was enhanced by serum exposure and this is shown in Figure [2A](#F2){ref-type="fig"}, representing a Northern blot utilizing poly-A selected RNA, hybridized with an 18S specific probe. By 5 minutes the intensity of the 18S band was more than tripled but back to baseline levels at 15 minutes. UBI4was also up-regulated by serum but its intensity remained the same at 15 minutes, while the 18S band returned to baseline assuring us that the 18S enhancement was not as a result from an error in RNA loading. We previously found ACT1\[[@B4]\] to show the same pattern, suggesting that serum exposure may also up-regulate constitutive genes. As a control for temperature, we also exposed yeast to YPD at 37°C and to serum at 30°C. There was no increase in polyadenylation in YPD at 37°C and there was increase with serum exposure at 30°C (data not shown) indicating that serum caused this increase. Estimates from phoshporimager data indicate that at baseline in YPD less than 1% of the rRNA is polyadenylated (data not shown). Polyadenylation of rRNA in Saccharomyces cerevisiaehas recently been described in similar amounts \[[@B6]\]. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Upregulation of 18S rRNA polyadenylation by serum exposure.Northern blot of poly-A selected RNA hybridized by 18S and UB14 specific probes (A). (a) is RNA derived from cells grown in YEPD at 30°C, (b) is RNA from yeast in serum for 5 minutes at 37°C, and (c) is RNA from yeast in serum for 15 minutes at the same temperature. (B) represents the quantitative phosphorimager data of the Northern shown in (A). Small letters (a, b, c) are the same as in (A). ::: ![](1471-2199-5-17-2) ::: To further confirm that the 18S increase from time zero to 5 minutes is real, we performed real-time PCR reactions and the results are shown in Figure [3](#F3){ref-type="fig"}. As can be seen, amplification can be detected 10 cycles earlier when the template is derived from organisms exposed to serum, confirming the increased amount of 18S in the starting material. 5S is detected at the same cycle whether exposed to serum or not. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Real-time PCR confirmation of 18S rRNA polyadenylation upregulation by serum.Real-time PCR reactions represented as PCR baseline subtracted relative fluorescence units (RFU) versus cycle number plots. Triangles represent reactions with 18S specific primers and circles represent reactions with 5S specific primers. ::: ![](1471-2199-5-17-3) ::: The 18S band in Figure [2A](#F2){ref-type="fig"}, lane b is slightly up-shifted as compared to lanes a and c, suggesting that likely one or more molecules, whose polyadenylation sites are downstream from the reported maturation sites are up-regulated by serum. This is consistent with our findings above, that 3 of the 4 polyadenylation sites located in that region were from serum exposed yeast and it suggests that serum up-regulation of polyadenylation may be selective to these downstream sites. When the same filter was hybridized with the 5S specific probe no bands were detected, indicating that our 18S bands were not a result of rRNA contamination (data not shown). Figure [2B](#F2){ref-type="fig"} represents the phosphorimager generated counts of these bands confirming the visual results objectively. Northerns that included the polyA-minus fractions (data not shown) continue to show the 18S bands indicating that rRNA transcripts with and without polyadenylated extensions are being produced. Discussion ========== These data indicate that 18S subunit mirrors the large 25S molecule as regards to polyadenylation and its response to serum, suggesting that this is not an incidental phenomenon. They do differ from the 25S subunit in that polyadenylation occurs both upstream and downstream to but not at the reported 3\'-end maturation site, while for the 25S subunit, the polyadenylation was found to be exclusively at the maturation site. Perhaps the 3\'-end of the 18S subunit plays an important role in the recognition of start sites on mRNA, and is vigorously protected from modifications. This would also suggest that polyadenylation of 18S has a role outside of the ribosome. One of the basic questions raised by our original data was whether these polyadenylated transcripts are products of Pol I and get polyadenylated following maturation cleavage or are newly transcribed by Pol II. These data do not resolve this question. While finding of multiple polyadenylation sites in 18S with most of them clearly not corresponding to a reported maturation site, might result from transcription by an enzyme other than Pol I, it is just as likely that they may represent inappropriate cleavage by the ribosomal RNA processing apparatus and these products are being readied for degradation by polyadenylation. The recent report of polyadenylation in Saccharomyces cerevisiaethat was found to be increased in mutants lacking the degradative function Rrp6p \[[@B6]\] favors the latter scenario. RNA polymerase switching from Pol I to Pol II for rRNA transcription has been described for Saccharomyces cerevisiae\[[@B7]\] in cells where the gene for one of the components of the Pol I transcription factor UAF (upstream activation factor) was mutated. These mutants gave rise to isolates that were utilizing Pol II for their rRNA transcription and this newly switched-on state was heritable even through meiosis. These data indicate that S. cerevisiaehas the inherent capacity to utilize Pol II for rRNA transcription but that this capacity is suppressed by a mechanism that includes UAF. These mutants though, switched to Pol II transcription exclusively. Our data with C. albicansdiffers in that both polyadenylated and non-polyadenylated forms are produced simultaneously. Conrad-Webb and Butow \[[@B8]\] have described rRNA transcripts of various lengths that were polyadenylated, produced by a respiratory-deficient isolate of S. cerevisiae. The template utilized by this strain was an episomal copy of ribosomal DNA that contained a Pol II promoter sequence overlapping with the Pol I promoter. Recently, circular and linear rDNA plasmids have been reported in C. albicans\[[@B9]\] for the first time. Thus it is possible that one of these episomal elements also contains Pol II promoters allowing it to function as the template for polyadenylated forms of rRNAs. With our findings that polyadenylation also involves 18S, such Pol II promoters would have to be present for both subunit genes making Pol II role less likely. Polyadenylation of a small percentage of total RNA in Escherichia colihas been reported \[[@B10],[@B11]\] including rRNA and this polyadenylation occurred even in wild type organisms. Hence it appears, that polyadenylation of these stable molecules occurs more widely as a biological phenomenon. The role of polyadenylation of rRNA in C. albicansis unknown. Open reading frame analysis of the 18S subunit indicates that translation into protein is unlikely, as it would result in peptides shorter than 40 amino acids. Multiple polyadenylation sites upstream and downstream from the reported maturation site suggest, that these may be inappropriately processed molecules that are being readied for degradation, though one of the downstream sites may be an A2 processing site. The mere up-regulation of polyadenylation by serum prior to germination also does not indicate a role in hypha formation as other genes such as ubiquitin and actin also respond similarly. There are aspects to our new 18S and our previous 25S data that leave the possibility for a role in germination open. These include the transient nature of this up-regulation for both subunits just prior to germination and the possible selective nature of this process involving 18S. Conclusions =========== The ribosome is central to cellular function and the RNA component of this organelle assumes critical structural and catalytic roles. Our initial unexpected finding of polyadenylation of a portion of the large rRNA subunit is now extended to the small subunit. That this modification also involves the other major component of the ribosome points to a biological role for this process. The fact that the transient up-regulation of RNA polyadenylation from both subunits just precedes the phenotypic expression of germination, suggests a possible role in regulating Candida albicans\'polymorphic behavior. Methods ======= Organism and germination conditions ----------------------------------- Candida albicansSC5314 (obtained from W. Fonzi) \[[@B12]\] was grown in YPD medium (1%, w/v, yeast extract; 2%, w/v, peptone; 2%, w/v dextrose) at 30°C. Heat inactivated (56°C for 30 minutes) fetal bovine serum (FBS) (10%, v/v in H~2~O) was utilized to induce germination. Yeast cells were grown overnight in YPD at 30°C, harvested by centrifugation, washed once in H~2~O and transferred to FBS pre-heated to 37°C at 1--5 × 10^6^cells ml^-1^ RNA isolation ------------- RNA from cells at various growth conditions was obtained as follows. Incubating mixtures were rapidly cooled in an ice-water bath and were thereafter centrifuged at 4°C and washed with ice cold water once. Cell walls were digested by suspending the pellet in a buffer containing 1 M sorbitol, 0.1 M EDTA, 0.1%, v/v, β-mercaptoethanol and 100 U ml^-1^lyticase (ICN), in a volume 1/5 that of the volume of the of the original cell culture. The digestion proceeded for 10--20 minutes at 30°C. Adequacy of the digestion was monitored by testing a small drop of cell suspension in SDS for viscosity \[[@B13]\]. Mixtures were centrifuged at 800 × g and RNA was isolated by using the QIAGEN total RNA kit, following the manufacturer\'s protocol. RNA was precipitated with isopropanol, and either used immediately or stored in ethanol at -20°C. Polyadenylated RNA selection was carried out by following the Oligotex mRNA kit protocol (Qiagen). Northern blot analysis ---------------------- Samples of 50 ng of poly(A) RNA were electrophoresed on 1.2% formaldehyde agarose gel blotted to nylon membranes. A 262 bp long fragment of 18S was generated from reverse transcribed total RNA with the primers 5\'-TCGATGGAAGTTTGAGGCAA-3\' (P1) and 5\'-ATTCAATCGGTAGTAGCGACGGGC-3\' (P2) based on the previously published sequence (Barnes et al., 1991). After cloning into pCR2.1 (Invitrogen), and sequencing (T7 Sequenase 2.0 kit, Amersham) to confirm that it was 18S specific, the insert was released by EcoRI digestion and gel purification, and was labeled with P^32^by using random priming. As a control a 499 bp UB14probe was also generated with the primers 5\'-GAAGTCGAATCTTCTGACACCATCG-3\' (P3) and 5\'-TGGTGGAATACCTTCTTTGTCTTGG-3\' (P4). The primer design was based on the UB14sequence (Accession No. Z54197) reported by the Stanford Candida Genome project\'s World Wide Web site \[[@B14]\]. This amplified product was also cloned and sequenced to confirm its specificity. To assess the quality of our RNA, a 5S specific probe was generated with the primers 5\'-GGTTGCGGCCATATCTAGCAGAA-3\'(P5) and 5\'-AGATTGCAGCACAATAGTTTCGC-3\' (P6). These primers were based on the reported sequence of the 5S component of Candida albicansrRNA \[[@B15]\]. Phosphorimaging volume report analysis (Molecular Dynamics, Sunnyvale, CA) has been employed to quantify objectively the intensity of individual bands. We estimated the percentage of polyadenylated 18S, from phosphorimaging data derived from Northern blots comparing total RNA and poly-A selected RNA (derived from the same amount of total RNA). Real-time PCR analysis ---------------------- To confirm that the polyadenylated form of 18S is increased by five minutes, total RNA was predigested with RNase-free DNase I (New England Biolabs), then reverse transcribed with Superscript II RT (Invitrogen) utilizing an anchored polyT primer 5\'-AATTCGGCGAGCTCCGCGGCCGCGTTTTTTTTTTTT-3\' (DT12) to generate cDNAs from polyadenylated RNA molecules. The same reaction also contained the primer P6, specific for 5S rRNA subunit that does not get polyadenylated \[[@B15]\]. Using these cDNA templates, a 63 nt long sequence specific for 18S at positions 1585--1648 \[[@B5]\] was amplified by primers P2 and 5\'-TCAGCTTGCGTTGATTACGTCC-3\' (P7). In a separate reaction, primers P5 and P6 were used to amplify 5S. To insure that there was no genomic DNA contamination, we carried out PCR reactions utilizing primer pairs P2-P7 and P5-P6 on predigested total RNA that was not reverse transcribed and no products were generated (data not shown). Real time PCR reactions were carried out with iQsybr Green Supermix(Bio-Rad) as source of fluorescence, utilizing an iCycler (Bio-Rad) thermocycler. The cycling settings used were; initial denaturation for 3 min at 95°C followed by 40 cycles each consisting of 30s denaturation at 95°C, 30s primer annealing at 55°C and 30s extension at 72°C. Data were analyzed using iCycler iQ version 3.0a. Amplification of poly-A extended 18S rRNA ----------------------------------------- To identify polyadenylation positions involving 18S, total RNA was heated at 70°C for 5 minutes to minimize secondary structures. Reverse transcription was done with the anchored polyT primer DT12, utilizing Superscript II RT (Invitrogen). PCR products were amplified with primers DT12 and P1 which is situated 398 nucleotides from the reported 3\' end of the molecule using HotStarTaq (Qiagen). To show the presence of a mature sized polyadenylated molecule, amplification was also carried out with DT12 and 5\'-TATCTGGTTGATCCTGCCAGTAGTC-3\' (P8) situated at the 5\' end of 18S using FailSafe PCR System (Epicentre). Amplified products were subcloned into pCR2.1 (Invitrogen), a number of clones were picked and sequenced (T7 Sequenase 2.0 kit, Amersham). For the full-length clone, only parts of the 5\' and 3\' ends were sequenced. Authors\' contributions ======================= HL and CPW carried out and analyzed experiments. JF conceived of the study, designed experiments, analyzed data and wrote manuscript. All authors have read and agree with final manuscript. Acknowledgements ================ We thank P. Masson for assistance with graphics.	PubMed Central	2024-06-05T03:55:48.426308	2004-10-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522811/", "journal": "BMC Mol Biol. 2004 Oct 4; 5:17", "authors": [ { "first": "Jacob", "last": "Fleischmann" }, { "first": "Hong", "last": "Liu" }, { "first": "Chieh-Pin", "last": "Wu" } ] }
PMC522812	Background ========== In humans, goal-directed movements to an object in space are improved by foveal vision, namely by gaze anchoring on the object \[[@B1]\]. Since motor and visual information are encoded in different reference frames, accurate reaching and grasping movements in space require ongoing registration and coordinate transformation of visual percepts with gaze and limb positioning. One essential transformation is to convert the retinal image from eye-centered coordinates in a target location with respect to the head by taking into account the position of the eyes in the orbit \[[@B2]\]. Single-cell recordings in monkeys revealed that the neural substrate of the visual-to-motor coordinate transformations is a change in the visual or motor response properties according to gaze position in extrastriate visual, parietal, oculomotor, premotor and motor areas \[for review see \[[@B3],[@B4]\]\]. Functional magnetic resonance imaging (fMRI) studies have localized human homologues of such monkey areas and have showed that eye position signals modulate activity of extrastriate visual areas \[[@B5]\] and the parieto-frontal network related to hand-arm movements \[[@B6],[@B7]\]. Less is known at the earliest stage of visual processing, namely in the primary visual area (V1). Previous electrophysiological \[[@B8]-[@B10]\] and modeling \[[@B11]\] studies described eye position dependent activity in V1 neurons to a lesser extent than that reported in parietal and premotor cortex but consistent with the idea that both retinal and eye position signals may also converge at early vision stage. However, the direct influence of eye position on V1-related activity in humans has not been investigated. Consequently, we used a specific experimental design based on pattern-onset visual evoked potentials (VEP) recordings to study the effect of eye position exclusively on V1 activity in humans. Thus, we focused our investigation on the first major VEP component C1, obtained using pattern-onset stimulation, because its distribution over the scalp and its retinotopic properties indicate an origin from the calcarine fissure that is V1. This issue has been the conclusion of all previous studies over the past 10 years regarding the cortical visual areas that generate the early components of pattern-onset VEP \[for review see \[[@B12]\]\]. The most recent reports even demonstrated in individual subjects a close anatomical correspondence between modelled dipoles for the C1 component and sites of activation in the calcarine fissure obtained in fMRI in response to the same visual stimuli \[[@B12]-[@B14]\]. The purpose of the present report was therefore to investigate the eye position-dependent activity of V1 in humans by characterizing the early C1 component of VEP and testing its properties at different eye positions. Results and discussion ====================== The C1 component reverses classically in polarity for upper vs.lower visual field stimulation \[for review see \[[@B12]\]\]. Consequently, we first characterized the C1 component over the 20 subjects by observing its polarity inversion for a central eye position. Figure [1B](#F1){ref-type="fig"} illustrates the representative polarity inversion of the C1 component on the grand averaged of VEP over the 20 subjects and in response to stimuli in the upper and lower quadrants of the right visual fields. Equivalent VEP and C1 polarity inversion were obtained in response to stimuli in the upper and lower quadrants of the left visual fields. The polarity inversion of the C1 components on the grand average VEPs in response to upper and lower hemifield stimuli were described previously for occipito-parietal sites using a 10--20 system montage with 62 scalp sites (see box included in Figure [1](#F1){ref-type="fig"}). It is noteworthy that a similar polarity inversion was measured in our study using only two occipital intermediate sites (IN3 and IN4) of the modified 10--20-system montage (see Methods for details). Once the C1 component was characterized for both upper and lower quadrant visual fields, the effect of eye position on C1 amplitude was measured only for both left and right lower quadrants visual field that is for the most salient C1 component which we observed. Mean peak latencies of C1 was calculated in each subject in response to both left and right lower visual quadrants and for both IN3 and IN4 recording sites. They ranged between 94.8 ms and 98.0 ms that are consistent with the C1 latency range previously observed in numerous studies \[[@B12],[@B15]\]. For each subject, the C1 amplitude for five different eye positions (0°, 10° and 20° both left- and rightward) was then measured at these mean latencies for each lower quadrant and each recording site and, with respect to a 80 ms pre-stimulus baseline. We observed that checkerboard presented in the lower visual field elicited VEPs with modulated amplitudes of the C1 component in respect of eye position. Grand averaged VEPs over the 20 subjects in response to flashed stimuli in the right lower quadrant of the visual field for three different eye positions (0° and 20° both left- and rightward) are shown in Figure [2A](#F2){ref-type="fig"}. It is noteworthy that the eye position-dependent modulation of C1 amplitude was observed at the IN3 recording site for a 20° rightward eye position (Figure [2A](#F2){ref-type="fig"}, red trace) and at the IN4 recording site with a 20° leftward eye position (Figure [2A](#F2){ref-type="fig"}, green trace). In other words, the eye position effect was observed at the occipital recording site contralateral to the direction of the eye deviation. Equivalent VEPs but reverse eye-position effects were obtained in response to left field stimuli. A one-way Friedman repeated measures analysis of variance was conducted for each recording site and for each visual stimulation revealing a significant main effect of eye position on the amplitude of the C1 component. Complete data obtained at both recording sites, for each lower visual field are shown in Figure [2B](#F2){ref-type="fig"}. We chose to represent these data using relative amplitude measures that are amplitude for deviated eye position subtracted by the amplitude for central eye position. It means that amplitude zero corresponds to the central eye position situation. A post-hoc Dunnett\'s test (p \< 0.05) using 0° as reference showed that the amplitude of the C1 component measured contralaterally to the deviated eye position was significantly different from 0° except for one situation (Figure [2B](#F2){ref-type="fig"}, white histograms). Conversely, the amplitude of the C1 component recorded ipsilaterally to the deviated eye position was not significantly different from 0° excepted for one situation (Figure [2B](#F2){ref-type="fig"}, grey histograms). Note that no parametric effect was observed for the C1 amplitude between 10° and 20° of eye eccentricity (Student\'s t-test, p \< 0.05). The overall result of the present study revealed that the amplitude of first major component of VEP elicited by checkerboard (C1) is modulated by the eye position. The previous data linking the C1 component to a striate cortex generator, namely V1 activity (see introduction), led us to suggest that eye position influences the earliest cortical stage of visual processing. One may first suggest that the present results may be explained by the difficulty of maintaining eccentric fixation, which may have altered the pattern of fixational eye movements, such as microsaccades \[fast, conjugate jerks, smaller than 1/3°, see \[[@B16]\]\]. Recent studies have shown that microsaccades modulated neural activity in V1 \[[@B17],[@B18]\], but to our knowledge no study has examined the effect of maintaining eccentric fixation on the occurrence of microsaccades. In the present study, we were able to track spontaneous saccadic eye movements superior to 1°. A one-way Friedman repeated measures analysis of variance for each visual stimulation revealed no significant main effect of eye position on the amplitude (p = 0.35) and on the frequency (p = 0.45) of saccades superior to 1°. We cannot rule out quantitatively a possible role for microsaccades in the eye position-dependent V1 activity. Regardless of such putative effects, however, one may argue that in case of an increase of the number of microsaccades related to eccentric fixation, a similar effect in terms of magnitude would be observed for both left- and rightward deviation. This was not the case in the present study. The question also arises if the effects of changing C1 amplitude may be due to oculomotor signals and/or retinal disparity that is to the difference in the position of the visual stimulus on each retina related to relative monitor distance. The absence of any parametric effect for the C1 amplitude between 10° and 20° of eye eccentricity may indicate that eye position effects are not due to the difference in the horizontal retinal disparity, but one may argue that a putative relationship between the horizontal disparity and the C1 amplitude is not linear. We evaluated the difference of both horizontal and vertical disparity between the different eye\'s deviations in our study (see Methods for details and Figure [3](#F3){ref-type="fig"}). Figure [3B](#F3){ref-type="fig"} shows that the difference between the magnitude of disparity for the central eye position and each deviated eye condition depends on both eye deviation and the distance of the visual stimulus from the fixation point. Interestingly, such a difference in terms of horizontal and vertical disparity does not depend on the stimulated quadrant visual field in the range of the present checkerboard\'s width. In other words, the magnitude of the relative disparity for a given distance from the fixation point and for a given deviated eye condition is similar for both left and right visual field stimulation. Since we observed that the C1 amplitude varied inversely in function of the hemifield stimulation (Figure [2](#F2){ref-type="fig"}), the eye-position effect observed on the C1 amplitude cannot be simply related to a horizontal and/or vertical disparity effect. Both direct and indirect arguments also suggest that variations in the C1 amplitude are not due to attention. Firstly, the subjects were instructed to keep firmly visual attention on the fixation point suggesting that the potential degree of attention required fixating binocularly the red fixation dot was similar across the different deviated eye positions. Secondly, numerous recent studies gave impetus to an emerging view that V1 activity may be modulated by attention through delayed feedback signals (160--260 ms) from extrastriate and/or oculomotor structures while the initial C1 sensory response (50--90 ms) was not modulated by attention \[[@B13]-[@B15],[@B19]-[@B22]\]. Our present findings, together with those aforementioned, allow considering that both eye position and attention-related signals may affect the early stage of visual processing in different manner. The former may comes from extraocular muscle afferents and/or corollary discharges while the later is considered as a late top-down process \[[@B14],[@B20]\]. Finally, Trotter et al. (1999) have shown that an eye position signal (extraretinal signal) is involved in the neural modulation process dealing with the eye position-dependent visual response observed in area V1 of behaving monkeys. Oculomotor signals coming from extraocular muscle afferents and/or corollary discharges are considered as the substrate of such an eye position signal and have been previously described in V1 \[[@B23],[@B24]\]. Conclusions =========== The present results and previous works obtained from neural recordings in monkeys indicate that changes in eye position can modify response properties in V1 that is at the earliest cortical stage of visual processing. Among the visuo-motor processing allowing accurate reaching and grasping movements in space on the basis of the image seen by the retina, the primary visual area may be therefore one of the first cortical relay to convert the image in eye-centered coordinates into a target location by taking into account the position of the eyes in the orbit. It endorses some recent arguments pointing out that V1 could no longer be considered only in relation to the pattern of light falling on the retina but appears to be a cortical area in which contextual influences take place too \[[@B9]\]. Methods ======= Subjects -------- Twenty right-handed healthy volunteers with normal visual acuity (range age 19--29 years, 9 males) participated in VEP recordings after they provided their written informed consent. The study was approved by the Basse-Normandie ethics committee (Caen, France). Experimental design ------------------- The subjects were seated on a swivel armchair with their head stabilized with a headrest. An experimenter slowly rotated the armchair and locked it in one of the five different angles: 0°, 10° and 20° both left- and rightward from the center of the monitor leading to five different eye positions (Figure [1A](#F1){ref-type="fig"}). The vertical swivel axis passed through the base of the nose between both eyes. Such a passive vertical movement of rotation of the whole body stimulated only the horizontal semicircular canals of the vestibular system with a time constant around 15 sec \[[@B25]\]. The VEP recording was started at least 1 minute after the rotational movement in order to stay away from the influence of the passive whole body rotational movement of the subjects. The subjects had to fixate binocularly a red fixation dot continuously visible in the center of the display as stimulus was flashed in 1 of the 4 quadrants of the visual field. Upper quadrants were used only in the case of central eye position in order to observe the inversion of polarity of the C1 component that allowed its detection on VEP recordings. The stimulus consisted of a black and white rectangular checkerboard (12 × 9° of visual angle, 0.6 cycle.deg^-1^of spatial frequency) flashed against a black background (ISI = 500--1000 ms) and delivered by a visual stimulator (Nicolet, Madison, USA). The subjects were instructed to fixate continuously the fixation dot and to keep their attention on it, for each quadrant of the visual field so that the projection of the stimulus on the retina was equivalent whatever was the deviated eye position. The edges of the monitor and the space up to 1 meter around the monitor were masked with an opaque black sheet preventing any cue perception in the room except the flashing checkerboard. VEP recordings -------------- With respect to the purpose of the present study, we recorded VEP from the scalp using the two occipital intermediate sites (IN3 and IN4) of the modified 10--20-system montage \[[@B26]\] with both left and right mastoids serving as reference. The VEP from each site was recorded (Vicking, Nicolet, Madison, USA) at a sampling rate of 2500 Hz (0.1--100 Hz of band-pass filter with a 50 Hz notch filter). Prior to averaging, artifact rejection was performed to discard epochs with eye blinks. A total of 200 non-rejected epochs was averaged for each recording. Both horizontal and vertical eye movements were also monitored in each subject and during all VEP recordings, with EOG electrodes placed around the orbit of the right eye. The EOG system had a resolution superior to 1° of visual angle. All EOG records were analyzed by computer, using a dedicated software (SAMO) \[[@B27]\] which detects saccadic components and quantifies the amplitude and frequency of the spontaneous saccadic eye movements. Measurement of the disparity ---------------------------- We evaluated the difference of disparity between the different eye\'s deviations (Figure [3](#F3){ref-type="fig"}). The difference in retinal angles (β - α) defines the magnitude of disparity, classically designated (η) \[[@B28]\]. Such a disparity depends on the position of the head\'s subject relative to the screen, designated (θ), on the interocular distance (a), and on the distance (d) between the middle of the cyclopean axis (c) and the fixation point (e1). We evaluated the magnitude of the disparity for any point (e2) of the checkerboard (Figure [3A](#F3){ref-type="fig"}). Using the dot product, for the left eye, α can be expressed as following: a1e1.a1e2 = \\|a1e1\\| \\|a1e2\\| cos(α) (1) Expressing of the coordinates of the vectors a1e1 and a1e2 in the canonical reference (x,y,z) centred on the cyclopean axis (c): a1e1.a1e2 = a1e1~x~a1e2~x~+ a1e1~y~a1e2~y~+ a1e1~z~a1e2~z~ (2) and combining both (1) and (2) led to: cos(α) = (a/2 cos(θ))(e2~x~+a/2 cos(θ)) + (d +a/2sin(θ))^2^/ \\|a1e1\\| \\|a1e2\\| (3) Similarly, for the right eye, we obtained: cos (β) = (-a/2cos(θ))(e2~x~-a/2cos(θ)) + (d -a/2sin(θ))^2^/ \\|a1e1\\| \\|a1e2\\| (4) Applying the numeric values defined in our study (d = 1 m, a = 8 cm), and combining both (3) and (4) allows to estimate the magnitude of the disparity (η = β-α) in function of the distance from the fixation point for each given eye\'s deviation. Therefore, we plotted the magnitude of disparity with each distance (e) and for each eye\'s deviation \[-20°, -10°, 10°, 20°\] normalized by the disparity calculated for the central eye position (θ = 0°, Figure [3B](#F3){ref-type="fig"}). List of abbreviations used ========================== C1 : early component of visual evoked potential fMRI : functional magnetic resonance imaging IN3 and IN4 : occipital intermediate recording sites ISI : inter-stimulus interval V1 : primary visual area VEP : visual evoked potential Authors\' contributions ======================= All authors designed the study. FA and OE carried out the experiments and generated the method for data analysis. OE performed the statistical analysis and generated the method for estimation of the magnitude of disparity. FA and LP wrote the first versions of the manuscript. All authors read, discussed and approved the final version of the manuscript. Acknowledgements ================ The authors are deeply indebted to Lydie Gabrel and the staff of the Service d\'Explorations Fonctionnelles du Système Nerveux (CHU Caen, France) for their help in data acquisition and to Michael Beauchamp, Marc Joliot, Bernard Mazoyer, Emmanuel Mellet, and Nathalie Tzourio-Mazoyer for their thoughtful comments on the first versions of the manuscript. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A. Experimental design. See methods for details. B. Polarity inversion of the C1 component observed on the grand averaged VEP over the 20 subjects and in response to stimuli in the upper (in orange) and lower (in purple) quadrants of the right visual fields. The present polarity inversion was observed on both IN3 and IN4 intermediate occipital sites. The box adapted from Di Russo et al (2002) represents the polarity inversion of the C1 components on the grand average VEPs in response to upper (solid line) and lower (dashed line) hemifield stimuli. In this study, waveforms were collapsed across VEPs to left and right hemifield stimuli and were plotted separately for scalp sites contralateral (left) and ipsilateral (right) to the side of the stimulation. The polarity inversion was observed prominently on occipito-parietal sites PO3/4 using a 10--10-system montage. ::: ![](1471-2202-5-35-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### A. Example of modulation of the C1 amplitude observed for both left- and right-ocular deviations of 20°. The grand averaged VEP over the 20 subjects and in response to stimuli in the lower quadrants of the right visual field is presented at each lateral site (IN3, IN4) for central eye position (blue) and, 20° leftward (green) and 20° rightward (red) eye positions. B. Comparison of the difference of C1 amplitudes between each deviated eye position and the central eye position for both left and right lower quadrant field. For each subject, the C1 amplitude calculated for each eye position was subtracted from the C1 amplitude measured for the central eye position. Vertical bars represent the standard error of the mean. ::: ![](1471-2202-5-35-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### A. Schematic experimental design. The cyclopean axis (c) is defined by the middle of both left and right eye rotation axis, a1 and a2 respectively. Both a1 and a2 also designates the interocular distance (a). θ corresponds to the subject\'s head position relative to the screen axis. Along the horizontal axis crossing the screen, (e1) corresponds to the centre of the screen, namely the fixation point, and (e2) represents a given point on the screen. Both α and β correspond to the retinal angles of the segment \[e1e2\] seen by the left and the right eye, respectively. The distance (δ) between the middle of the cyclopean axis (c) and the fixation point (ε1), in other words the distance between the screen and the subject was 1 m in the present study. B. Plot of the magnitude of disparity for each point of the checkerboard and for each eye\'s deviation \[-20°, -10°, 10°, 20°\] normalized by the disparity calculated for the central eye position (θ = 0°, see methods for details of the calculation). ::: ![](1471-2202-5-35-3) :::	PubMed Central	2024-06-05T03:55:48.427955	2004-9-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522812/", "journal": "BMC Neurosci. 2004 Sep 19; 5:35", "authors": [ { "first": "Frédéric", "last": "Andersson" }, { "first": "Olivier", "last": "Etard" }, { "first": "Pierre", "last": "Denise" }, { "first": "Laurent", "last": "Petit" } ] }
PMC522813	Background ========== The gastrointestinal tract is the initial barrier to dietary constituents and is important in regulating nutrient entry, as well as keeping non-nutrients out. Additionally, the hepatobiliary system acts as an additional filter to rapidly excrete such non-nutrients into bile, thus keeping the net retention of these potential toxins low. Mammals have evolved many mechanisms in the gastrointestinal tract to select out usable dietary constituents from those that maybe potentially toxic to the body. It is apparent that the ATP-binding cassette proteins (ABC proteins/transporters) are the machinery that mediate the ATP-dependent transport of a wide variety of substrates that range from xenobiotics to peptide fragments \[[@B1]\]. A subset of these ABC transporters, located in the canalicular membranes of mammalian liver, play key roles in bile formation and detoxification \[[@B1]-[@B3]\]. One of these processes involves the regulation of sterol entry and excretion. Whole body cholesterol homeostasis is a tightly regulated process, involving dietary absorption, de novosynthesis and hepatobiliary secretion. Sitosterolemia, a rare autosomal recessive disorder of sterol metabolism results in the disruption of dietary sterol entry and hepatobiliary sterol secretion \[[@B4]-[@B6]\]. Under normal circumstances, our diets contain equal amounts of plant sterols and cholesterol, but the plant sterols are specifically excluded from our bodies and only regulated amounts of cholesterol are retained. In Sitosterolemia, this exclusion is defective resulting in the retention of non-cholesterol sterols. Mutations in either, but not both, of two ABC transporters, ABCG5 and ABCG8, encoded by a single locus, STSL, are known to cause this disease \[[@B7]-[@B9]\]. Based upon the genetics, as well as in vitroand in vivodata, these \'half-transporters\' are proposed to function as obligate heterodimers. In vitroexperiments have shown that both proteins are needed to be co-expressed for apical expression and that these may function as mutual chaperones in the ER for maturation \[[@B10],[@B11]\]. In vivoexperiments in mice have not been consistent. Using the Abcg5/Abcg8 double knockout mice, Graf et al has shown that by inoculating them with adenoviral constructs for Abcg5 and Abcg8 that both are required for expression of both proteins \[[@B11]\]. Additionally, Plosch et aland our group have constructed mouse models deficient in either Abcg5 \[[@B12]\] or Abcg8 \[[@B13]\] that show different biliary physiology than that of the Abcg5/Abcg8 double knockout mice. This suggests these proteins may have independent function(s) in addition to their function as heterodimers. However, to date, no reports of characterization and localization of the human proteins have been reported. In this report, we examined the location of these two proteins using cellular fraction and immunohistochemical analyses of human liver, gallbladder and small intestine. We found a general concordance of co-expression of both proteins, but we also noted that ABCG5 and ABCG8 could be found in plasma membranes, as well as in intracellular membrane locations independent of each other. Additionally, deglycosylation of human liver membranes with peptidyl N-glycosidases did not alter the mobility of the proteins after SDS-PAGE, suggesting that these proteins may not be glycosylated in human liver. This differential localization suggests that perhaps ABCG5 and ABCG8 may have functions independent of each other, as well as functioning as heterodimers. Methods ======= Tissue aquistion. Human liver donated for transplantation, but deemed unsuitable for transplantation on inspection by the transplant service (usually based upon a \'fatty\' appearance) was obtained in accordance with IRB approval. As soon as the liver was deemed unsuitable (typically less than 10 h following harvesting) pieces were either snap-frozen in liquid nitrogen and stored at -80°C or liquid nitrogen until use, or placed in ice-cold 2-methylbutane and stored in liquid nitrogen. Samples from more than nine different donors were used in these studies. Additionally, human gallbladders and segments of proximal small intestine were obtained from patients under going either laproscopic cholecystectomy or pancreatoduodenectomy (Whipple procedure). These tissues were directly taken from the operating room in normal saline on ice to be processed directly for frozen sectioning. Antibodies ---------- Anti-membrin, anti-transferrin, and anti-calnexin antibodies were obtained from Stressgen (Victoria, BC Canada), anti-caveolin antibody from BD Bioscience (San Diego, CA, USA), anti-MDR1 that also detects MDR2/3 (C219) from Centocor Inc. (Malvern, PA, USA), anti-MRP2 (cMOAT) from Chemicon International (Temecula, CA, USA) and secondary antibodies were purchased from Jackson Immuno research (West Grove, PA, USA). Polyclonal rabbit anti-sera to human ABCG5 and ABCG8 peptides were generated in-house, using a 20-peptide immunogen from human ABCG5 (576--587, FQKYCSEILVVNEFYGNFTC, GenBank Accession number NP\_071881) and a 22-peptide sequence from human ABCG8 (608--629, SRRTYKMPLGNLTIAVSGDKIL, GenBank Accession number NP\_071882). The anti-sera were further purified using peptide affinity columns and stored at a concentration of 0.8 mg/ml in Immuno Pure Binding Buffer (Pierce, Rockford, IL, USA). For peptide blocking experiments the peptides were dissolved in DMSO (final concentration 30%), incubated with the corresponding peptide for 1 1/2 hours at 37°C then used for immunoblotting as described below. Membrane protein preparation ---------------------------- Crude total membrane isolation was carried out with minor modifications as previously described \[[@B14]\]. All the procedures were carried out at 4°C. Three grams of human liver were homogenized in homogenization buffer (5 mM Tris pH7.5, 250 mM sucrose, 1 mM PMSF, 20 μg/μl of leupeptin and 1 μg/μl of aprotinin) by applying 10 strokes with a dounce homogenizer. The homogenate was centrifuged at 1000 gfor 10 minutes, the pellet containing any undisrupted cells and nuclear debris were re-homogenized with one-half the initial volume of homogenization buffer, centrifuged at 1000 gfor 10 minutes and this process was repeated once more. Supernatants were pooled and subjected to centrifugation at 100,000 gfor 40 minutes. The resulting pellet, deemed the crude membrane fraction, was used as the starting material for Western blotting and fractionation experiments. Nycodenz gradient fractionation ------------------------------- Human liver crude total membrane proteins were re-suspended in 30% Nycodenz solution (Nycodenz in 5 mM Tris-HCl pH 7.5 and 1 mM EDTA). This suspension was loaded on top of a 40% Nycodenz solution cushion in an ultracentrifuge tube, overlaid by consecutive 23%, 20%, 15% and 10% Nycodenz solutions and subjected to centrifugation at 39,000 rpm for 16 hours at 4°C in a SW41 rotor (Beckman Instrument, Palo Alto, CA). After centrifugation, 800--1000 μl fractions were sequentially removed from the top, combined with two volumes of homogenization buffer (see above) and centrifuged at 39,000 rpm at 4°C for 40 minutes to remove the Nycodenz. The resulting pellets were re-suspended in buffer (25 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.1% Triton X-100 and 0.1% SDS, 1 mM PMSF, 20 μg/μl of leupeptin and 1 μg/μl of aprotinin), the protein content determined by the method of Lowry and fractions analysed by SDS-PAGE. Equal amounts of protein (25 μg) per lane were loaded. Sucrose gradient fractionation ------------------------------ The procedure for membrane fractionation was essentially as described for the Nycodenz fractionation, except for the homogenization buffer used (25 mM Tris pH6.8, 150 mM NaCl, 1 mM PMSF, 20 μg/μl of leupeptin and 1 μg/μl of aprotinin). The sucrose density gradient fractionation was modified as previously described \[[@B15]-[@B17]\]. Human liver crude membrane proteins were re-suspended in 1% Triton X100 buffer (25 mM Tris-HCl pH6.5, 150 mM NaCl, 1% Triton X100, 1 mM PMSF, 20 μg/μl of leupeptin and 1 μg/μl of aprotinin), adjusted to a final sucrose concentration of 40% and overlaid with a 15--30% linear sucrose gradient. The samples were subjected to centrifugation at 39,000 rpm for 16 hours at 4°C in a SW41 rotor (Beckman Instrument, Palo Alto, CA, USA) and fractions collected from the top as described above. The proteins from fractions 1--4 from top of the tube were precipitated with acetone because these fractions did not contain sufficient protein for direct analysis. After protein concentrations were determined, equal amounts of proteins (20 μg) from each fraction were resolved by SDS-PAGE. Immunoblotting -------------- Proteins resolved by SDS-PAGE were transferred onto nitrocellulose membranes. Membranes were then blocked for 1 hour in 5% dry milk in PBS-T (Phosphate Buffered saline and 0.1% Tween 20) and incubated with primary antibody against either ABCG5 or ABCG8 in 5% milk in PBS-T overnight at 4°C. Blots were washed three times for 5 minutes in TBS-T (Tris Buffered Saline/0.1% Tween-20) with 150 mM NaCl, incubated with goat-anti-rabbit conjugated HRP antibodies (1:10000 dilution), washed for three times 5 minutes and developed with Western Lightning^®^Chemiluminescence Reagent Plus (Perkin Elmer Life Sciences, Inc. Boston, MA, USA). Immunohistochemical analysis and microscopy ------------------------------------------- Snap-frozen liver, gallbladder and intestine tissues were used to cut 8 μm thick frozen sections, air-dried for 30 minutes onto glass slides and kept at -80°C until used. The slides were stained with hematoxylin, rinsed with PBS three times, fixed for 10 minutes with cooled methanol at -20°C and rinsed with PBS three times. The slides were treated with blocking solution (10% donkey serum in 0.1 M glycine/PBS) for 30 minutes at room temperature and incubated with primary antibody overnight at 4°C. The slides were washed with PBS and incubated with secondary antibody (goat-anti-rabbit conjugated with Cy3™ or rhodamine or FITC) for 20--30 minutes at room temperature, rinsed with PBS three times and examined under an Olympus BX-5 confocal microscope with Fluoview. Results ======= Identification of ABCG5 and ABCG8 in crude total membrane preparations of human liver ------------------------------------------------------------------------------------- Peptide antibodies were raised against human ABCG5 and human ABCG8 and affinity purified prior to use (see Methods). The immunogen peptides used for the antibodies were selected since they were sequences that lay outside of the predicted transmembrane domains and based upon antigenicity.Western blotting experiments (Figure [1A](#F1){ref-type="fig"}) showed that both anti-ABCG5 and anti-ABCG8 antibodies bound to \~75 kDa proteins in human liver crude membranes. Pre-immune sera did not detect the \~75 kDa expected bands. Pre-incubation of the immune antibodies with the peptides against which they were raised abolished specific binding (Figure [1B](#F1){ref-type="fig"}). For anti-ABCG5 5 μg of peptide was needed to block 1 μg of antibody and for anti-ABCG8 12.5 μg of peptide was needed to block 1 μg of antibody. Interestingly, a \~60 kDa band was detected using the anti-ABCG8 antibody whose signal is abolished when incubated with the peptide from which the antibody was raised (Figure [1B](#F1){ref-type="fig"} and [1C](#F1){ref-type="fig"}, arrow indicated band). The significance of this is unclear at present. These antibodies were also tested against mouse and rat liver membrane preparations and no significant cross-reactivity was detected except for faint bands seen with anti-ABCG5 in mouse liver samples (Figure [1A](#F1){ref-type="fig"}, tracks 3 and 4). No other bands were detected above the 150 kDa marker in all western blots. Interestingly, while these proteins are predicted to be N-glycosylated \[[@B8],[@B9]\], only single bands in the appropriate molecular weight range were detected and no higher molecular bands were observed. To investigate whether these proteins are glycosylated, crude membrane fractions were digested with EndoH, PNGase F and examined for alterations in gel migration by SDS-PAGE (Figure [1C](#F1){ref-type="fig"}). Although the mobility of a known glycoprotein, transferrin, was increased in the same fractions following deglycosylation, there was no change in the mobility of ABCG5 or ABCG8 (Figure [1C](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Immunoblotting analyses of ABCG5 and ABCG8 in human liver.Panel A shows the immunodetection of ABCG5 (tracks 1--6), ABCG8 (tracks 7--12) in membrane preparations from human liver (HL, tracks 1, 2, 7, 8) mouse liver (ML, tracks 3, 4, 9, 10) or rat liver (RL, tracks 5, 6, 11, 12). The anti-ABCG5 peptide antibody detected a faint mouse band, but no other specific binding was identified. Although the pre-immune sera detected bands in the rodent tissue samples, none were detected in human liver (tracks 13--16, MB, mouse brain). Specificity was further shown by pre-incubation of the antibodies with the peptides they were raised against (panel B). In the presence of the specific peptides, the 75 kDa bands are not detected in human liver microsomes. Panel C shows the results of deglycosylation of human total liver microsomes, probed with anti-ABCG5 (left hand panel), anti-ABCG8 (middle panel) or anti-transferrin (right hand panel). Aliquots from the same incubation were separated for all three western blots. Although ABCG5 and ABCG8 do not appear to have their SDS-PAGE mobility\'s altered by either EndoH or PNGase F treatment, that of transferrin in the same samples is clearly effected (see Text for discussion). Newly synthesized (sensitivity to EndoH), as well as mature forms of transferrin (resistant to EndoH, but fully sensitive to PNGase F) are present in the liver membrane preparations. ::: ![](1471-230X-4-21-1) ::: Localization of ABCG5 and ABCG8 by Nycodenz and Sucrose gradient fractionation of human liver --------------------------------------------------------------------------------------------- Crude total membrane proteins from human liver were fractionated by Nycodenz gradient centrifugation and examined for localization markers by western blot analyses (Figure [2A](#F2){ref-type="fig"}). After Nycodenz gradient centrifugation, ABCG5 (fractions 9--10) and ABCG8 (fractions 6--11) were found to have a broad range of distribution and appeared to be distributed in a pattern similar to calnexin (an ER membrane marker, fractions 5--10), Cytochrome C (a mitochondrial marker, fractions 4--9), transferrin (a plasma membrane marker, fractions 1--11), caveolin (fractions 6--10) and MDR1 (an apical membrane marker, fractions 4--10). ABCG5 and ABCG8 did not co-localize with cis-Golgi (Figure [2A](#F2){ref-type="fig"}) markers. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Subcellular localization of ABCG5 and ABCG8 in human liver.Panel A shows the Nycodenz gradient fractionation and panel B the Triton X-100/sucrose gradient fractionation. A representative result from each of these is shown. F1--F12 represents serial fractions collected from the top of the tube. Proteins from each of these fractions were separated by SGS-PAGE, western blotted and probed for the proteins as indicated on the figure. Calnexin (ER marker), membrin (Golgi marker) and caveolin (raft marker) did not co-localize with ABCG5/ABCG8 when both biochemical fractionation patterns are compared. However, MDR1 (apical membrane marker) and transferrin (plasma membrane marker) showed some consistency with ABCG5/ABCG8 co-localizations (tracks F9--10, panel A and F11, panel B). ::: ![](1471-230X-4-21-2) ::: To examine whether ABCG5 and ABCG8 were associated with membrane rafts, total membrane proteins from human liver were solubilized with ice-cold 1% Triton X-100 detergent and fractionated by sucrose density gradient centrifugation (Figure [2B](#F2){ref-type="fig"}). Fractionation resulted in two Triton X-100 insoluble complexes, as judged by the clarity of the gradient fractions. The first was found in the low-density range (15--20% sucrose, fractions 2--6, Figure [2B](#F2){ref-type="fig"}) and the second in the high-density range (40% sucrose, fractions 10--12, Figure [2B](#F2){ref-type="fig"}). Caveolin-rich fractions localized to the low-density range (Figure [2B](#F2){ref-type="fig"}, fractions 2--6). However, ABCG5 and ABCG8 were detected in the high-density fractions, F10--12, along with transferrin (fractions 5--12) and MDR1 (fractions 9--11). Calnexin or cytochrome C, under these conditions did not co-localize with either ABCG5 or ABCG8. Note that the majority of the ABCG5 and ABCG8 were present in the densest fractions, F11--12. Under these conditions, membrin, a cis-Golgi membrane marker, was also detected in fraction 12. Thus, ABCG5 and ABCG8 have significant overlap with each other suggesting co-localization. However, these proteins did not seem to co-localize with any specific membrane marker except transferrin, when two different methods of fractionation were utilized. Immunohistochemical localization of ABCG5 and ABCG8 in human liver ------------------------------------------------------------------ It has been shown previously that ABCG5 and ABCG8 are expressed only in the liver and intestine \[[@B8],[@B9]\]. To further characterize the location of ABCG5 and ABCG8 in human liver, immunohistochemical analyses were performed on frozen serial sections of human liver. Pre-immune sera were used as negative controls. The distribution of the two proteins appeared to be divergent not only histologically, but also at the cellular level. From a histological point of view, ABCG5 was detected along sinusoidal tracts (Figure [3A](#F3){ref-type="fig"}, upper panel) whereas ABCG8 was detected within the cells lining the bile ducts (Figure [3B](#F3){ref-type="fig"}, upper panel). At higher magnification, ABCG5 was detected along bile canaliculi and at the cellular level appeared to be an apically expressed (Figure [3A](#F3){ref-type="fig"}, lower panel). However, the distribution of ABCG8 at the cellular level appeared more diffuse consistent with plasma membrane expression and perhaps in intracellular membranes (Figure [3B](#F3){ref-type="fig"}, lower panel). Expression of ABCG5 within intracellular vesicular compartments could not be excluded by the techniques employed. To confirm an apical location of ABCG5 and ABCG8 immunohistochemical co-localization studies were carried out using an antibody against a known apical transporter (MRP2) in the liver. As shown in Figure [4](#F4){ref-type="fig"}, ABCG5 and ABCG8 have significant overlap in expression with MRP2 (panels A and B, respectively). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Immunolocalization of ABCG5 and ABCG8 in human liver sections. Panel A shows the staining pattern of ABCG5 and panel B that for ABCG8. The pre-immune controls for both antibodies are as marked and shown in the top right hand corners of each panel. The top panels of each section are at low magnification (bar is 50 μm) and the bottom panels at high magnification (10 μm). The images for ABCG5 and ABCG8 were visualised with red and green colors respectively using Adobe Photoshop (Adobe, Cupertino, CA). The left panels show hematoxylin stained phase contrast images and the middle panels show the fluorescence images after immune serum staining. The bottom right panel of each section shows the merged images of phase contrast and the fluorescence signals. ABCG5 was readily detectable in canalicular cells and at higher magnification seemed to be apical in expression (panel A). On the other hand, ABCG8 was more readily detectable in cells lining the bile ducts (panel B, top panels), as well as in canalicular cells; although its cellular expression appeared more diffuse (see Text for discussion). ::: ![](1471-230X-4-21-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Co-localization of ABCG5 and ABCG8 with MRP2 by immunohistochemistry.Liver sections were simultaneously incubated with either the combination of anti-ABCG5/MRP2 or anti-ABCG8/MRP2. Panel A shows staining patterns of MRP2 and ABCG5 independently (upper portion) then merged together (lower right) with what appears to be similar over-lapping expression patterns. Likewise MRP2 and ABCG8 expression patterns appear to over lap as well as shown in panel B. Bar is 20 μm. ::: ![](1471-230X-4-21-4) ::: Immunohistochemical localization of ABCG5 and ABCG8 in human gall bladder ------------------------------------------------------------------------- Serial sections of human gall bladder were incubated with each antibody and pre-immune serum was used as a negative control. Both ABCG5 and ABCG8 were detected in the epithelium of gall bladder mucosa (Figure [5A](#F5){ref-type="fig"} and [5B](#F5){ref-type="fig"}). At higher magnification, the cellular distribution of the signals detected was similar and showed a diffuse cytoplasmic distribution (Figure [5A](#F5){ref-type="fig"} and [5B](#F5){ref-type="fig"}). ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Immunolocalization of ABCG5 and ABCG8 in human gall bladder sections.Gall bladder surgical samples were stained for ABCG5 (panel A, red staining) or ABCG8 (panel B, green staining). The layout is as indicated for figure 3. Both ABCG5 and ABCG8 appeared to be confined to the epithelial cells lining the lumen and no significant staining in any of the deeper cell layers was detected. At the cellular level, both proteins seem to be diffusely expressed within the cell and at plasma membrane. A strict apical expression was not observed for either of these proteins. ::: ![](1471-230X-4-21-5) ::: Immunohistochemical localization of ABCG5 and ABCG8 in human small intestine ---------------------------------------------------------------------------- Both ABCG5 and ABCG8 were detected in the apical surfaces of the enterocytes in biopsy samples of the small intestine (Figure [6A](#F6){ref-type="fig"} and [6B](#F6){ref-type="fig"}, upper panels). However, at higher magnification the cellular distribution of ABCG5 appeared more diffuse (Figure [6A](#F6){ref-type="fig"}, lower panel) whereas ABCG8 was expressed apically (Figure [6B](#F6){ref-type="fig"}, lower panel). This divergent pattern was seen in all of the serial sections analyzed. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Immunolocalization of ABCG5 and ABCG8 in human intestinal sections.Human intestinal surgical samples were stained with ABCG5 (panel A, red staining) or ABCG8 (panel B, green staining). The layout is as indicated for figure 3. At the cellular level, ABCG5 was expressed in a more diffuse pattern (panel A, see bottom right hand panel). In contrast, ABCG8 was expressed in the apical surfaces of the enterocytes (panel B, bottom right hand panel). Both proteins seemed to be expressed only in the enterocytes lining the villi and no significant expression was detected in any of the other cell layers. ::: ![](1471-230X-4-21-6) ::: Discussion ========== In this study, we report the localization of ABCG5 and ABCG8 in human liver, gall bladder and intestine. Our studies showed that these proteins are highly specific in the cells they are expressed. In the liver, expression is seen in cells lining the hepatobiliary tracts, both hepatocytes and ductal cells. In the intestine, robust expression was seen only in the villus enterocyte layers. In the gall bladder, expression was confined to the epithelial cells lining the lumen. However, some differences in the distribution of ABCG5 and ABCG8 within these tissues were apparent. In the liver, ABCG8 was highly expressed in the hepatocytes lining the bile ducts, whereas ABCG5 was more robustly expressed in hepatocytes lining the cannaliculae. In fractionation studies, using two different methods of separation, the distribution of ABCG5 and ABCG8 was compatible with both proteins potentially acting as heterodimers. However, we also noted that there were fractions where only one of these proteins, but not the other was detected. This could be an artefact, with one antibody being a better reagent, or that this pattern could truly reflect that each of these proteins can also exist independently, perhaps as homodimers. Overall, the distribution patterns of ABCG5 and ABCG8 in these cellular fractionations were similar to that observed with the plasma membrane marker transferrin and apical membrane marker MDR1. Additionally, immunohistochemical analyses show that ABCG5 and ABCG8 are apically expressed in the liver, gall bladder and intestine. In the liver ABCG5 and ABCG8 also appear to co-localize with the known apical protein MRP2. This confirms previous data from Graf et al, using in vitroexpression in WIF-B cells \[[@B10]\], and support the contention that ABCG5 and ABCG8 are plasma membrane proteins. These data would suggest that expression of these proteins might not be wholly dependent upon mutual co-expression, as has been reported for the mouse and in in vitrostudies \[[@B10],[@B11]\]. However, one note of caution should be expressed. All of the liver samples analyzed were obtained because they were unsuitable for transplantation. Most livers were considered to be \'fatty\' livers. While these livers were not effectively diseased, that fact that they had fatty infiltrates may have influenced the normal expression of these two half-transports. Thus, confirmation in normal human liver samples will be needed, though this may not be feasible. With that reservation in mind, our data do have important implications for sterol trafficking in humans. Firstly, the relatively robust and highly specific expression of ABCG5 and ABCG8 in gall bladder epithelium confirms the important role of this organ in regulating biliary secretion. In addition to the production of bile by the liver, the gall bladder may be able to further regulate the sterol content of bile, via ABCG5/ABCG8 activity. A similar pattern of expression has been reported in canine gall bladder epithelial cell culture and these data confirm these findings in human gall bladder \[[@B18]\]. Secondly, the differences in expression patterns of ABCG5 and ABCG8 in liver, gall bladder and intestine, though subtle, seem to indicate the several important possibilities. It is not clear which organ is of paramount importance in the human in regulating non-cholesterol sterol retention. And it is not clear if ABCG5 and ABCG8 play a significant role in determining cholesterol entry at the intestinal level, though they seem to be implicated strongly as determining sterol excretion at the level of the hepato-biliary system. In mice deficient for Abcg5/Abcg8 or Abcg5, cholesterol absorption rates were not dramatically affected \[[@B12],[@B19],[@B20]\]. Hepatobiliary excretion of all sterols was significantly (but not completely) reduced. In mice singly deficient for Abcg5 or Abcg8, some differences have been reported \[[@B12],[@B13]\]. In Abcg5 KO mice, following LXR activation, sterol excretion in bile was comparable to wild-type mice, though it is not clear if this also restored plant sterol excretion. In contrast, although cholesterol absorption studies for Abcg8 KO mouse have not been reported, biliary excretion of cholesterol was dramatically reduced and no stimulation of excretion was observed after forced bile acid infusions. While all of these data have been reported using different protocols (LXR activation, bile acid infusions, or static gall bladder puncture), these data suggest that there exist other pathways for sterol trafficking in both liver and intestine. At present it would be speculative to assume that these \'other\' pathways involve ABCG5 or ABCG8 as homodimers, but this possibility is supported by the circumstantial evidence of separate patterns of expression of human ABCG5 and ABCG8 in these tissues. Finally, although rodent sterolins are glycosylated and in vitroglycosylation is readily demonstrable, human ABCG5 and ABCG8 did not appear to be glycosylated as judged by deglycosylation-migration assays. It is possible that this technique is insensitive and these proteins are glycosylated. Alternatively, it is possible that the antibodies we have raised only react with unglycosylated forms and thus fail to detect the glycosylated forms. With respect to the first issue, deglycosylation-migration has been demonstrated to detect mouse glycosylated proteins and since these proteins are highly conserved, this possibility seems remote. With respect to the second possibility, if our antibodies were exclusively detecting unglycosylated (and presumably immature forms), the apical patterns of expression of these proteins in both the liver and intestine would seem to suggest that these proteins seem to traffic to these specialized membranes normally. In the absence of an independent method, and the lack of a direct assay for function, whether these proteins form an active heterodimer can not be resolved at present. Conclusion ========== In summary, we report the first immunolocalization of ABCG5 and ABCG8 in human liver, gall bladder and intestine. Our data show that these proteins are located in membranes and can have an apical expression in all of these tissues. Biochemical, as well as immunolocalization studies show that while both proteins co-localize in general, they can also seem to have expression patterns that may be independent of each other. Competing interests =================== None declared. Authors\' contributions ======================= ELK and MHL performed the experiments, KDC and DBA provided the liver and surgical samples respectively SBP was responsible for supervision, data analyses and obtaining funding for these experiments. ELK and SBP wrote the paper. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-230X/4/21/prepub> Acknowledgements ================ Grants from the NIH, HL606162 (SBP) and by NIH Postdoctoral Training Grant T32 HL07260 (ELK) supported this work. We would like to thank Brenda Baldwin, FNP for her assistance in obtaining consent from donor patients.	PubMed Central	2024-06-05T03:55:48.430398	2004-9-21	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522813/", "journal": "BMC Gastroenterol. 2004 Sep 21; 4:21", "authors": [ { "first": "Eric L", "last": "Klett" }, { "first": "Mi-Hye", "last": "Lee" }, { "first": "David B", "last": "Adams" }, { "first": "Kenneth D", "last": "Chavin" }, { "first": "Shailendra B", "last": "Patel" } ] }
PMC522814	Background ========== Leptin, a 16-kilodalton protein, is involved in the regulation of food intake and body composition \[[@B1]\]. It was discovered in 1994 by Friedman et al. \[[@B2]\] and has been proposed to physiologically regulate body weight by suppressing appetite and increasing energy expenditure \[[@B1],[@B3],[@B4]\]. In normal humans, circulating level of leptin is higher in women than in men \[[@B3],[@B5]\]. Besides of gender dependency, circulating leptin levels correlate with the body fat mass (BFM) and body mass index (BMI) in healthy subjects \[[@B5]-[@B7]\]. Malnutrition is a common feature of cirrhotic patients \[[@B8]\]. A negative energy balance, and thus catabolism caused by energy expenditure is considered to be of pathophysiological relevance in cirrhosis \[[@B9]\]. Several studies have shown that circulating leptin levels are modestly elevated in patients with alcoholic cirrhosis, suggesting that leptin might be involved in the malnutrition of cirrhosis \[[@B10],[@B11]\]. While some studies have been supported these findings, others have reported low serum leptin levels in post-hepatitis cirrhotic patients \[[@B10],[@B12],[@B13]\]. In addition, nutritional status of cirrhotic cases represents a wide range in normal to severe malnutrition, connected with severity of the disease \[[@B8]\]. It appears that relationship of serum leptin levels and nutritional status in post-hepatitis cirrhosis has not been fully clarified yet. In this study, we investigated the relation between the severity of disease and serum leptin levels in post-hepatitis cirrhosis and the role of body composition, gender and viral aetiology of cirrhosis in this association. Methods ======= Thirty-five cases with post-hepatitis cirrhosis (17 male, 18 female; mean age: 51.5 ± 12) which were diagnosed on the basis of the clinical, laboratory, radiological, and/or histopathological findings, and 15 healthy controls (8 male, 7 female; mean age: 49.4 ± 8) were enrolled in this study. Cirrhotic cases were assigned into 3 groups on the basis of the Child-Pugh classification \[[@B14]\] as follows: Child A (n = 10), Child B (n = 14) and Child C (n = 11). Causative agents of cirrhosis were viral hepatitis B (n = 20) and hepatitis C (n = 15). As leptin is a gender dependent peptide, control and cirrhotic group were divided into two groups as male and female. Exclusion criteria were history of cancer, diabetes mellitus, and alcoholism, existence of pleural effusion, gastrointestinal bleeding, acute infection and renal failure, treatment with corticosteroids, immunosuppressive agents and oral contraceptive within the last 6 months. Control group consisted of healthy individuals with normal medical history, physical examination and blood biochemistry. None of them have had a restriction of diet for loosing weight during the last three months. Subjects who receive any medication have not been included into control group. The local human institutional review committee approved the study and written consents were received from all participants. Body composition such as BMI, skin fold thickness, body fat percentage (BFP), and BFM analysis was performed in both cirrhotic cases and controls. To avoid incorrect BMI determination and body composition analysis, cirrhotic cases with ascite and edema had been put on sodium restricted diet of 51 mmol per day and they were received diuretics (spiranolactone 100--200 mg and, if necessary, furosemide 40--80 mg per day) until ascite and edema have been resolved. Cirrhotic cases with refractory ascite unresponsive to therapy impaired renal function following diuretic therapy, and triceps skinfold thickness less than 10^th^percentile \[[@B15]\] were excluded. BMI was determined as the actual body weight relative to the square of the body height (BMI, kg/m^2^). Measurements of skin fold thickness were conducted at four different sites on the left side of the body (triceps, biceps, sub-scapular and supra-iliac) using a Holtain skinfold caliper (Holtain, Crosswell, Crymych, Dyfed, UK). All the measurements were made by the same physician (FB). Two measurements were made at each site and the average values were obtained. The BFP was calculated using the Jacksons\' formula \[[@B16]\]. BFM was calculated using BFP and body weight as kilogram. Diet containing 1 g/kg body weight of protein and 30 kcal/kg body weight of non-protein calories, was described to be consumed by both cirrhotic and controls for 2 weeks before serum leptin level measurement was performed. Blood samples were obtained in the morning following 12 hours of fasting, they were centrifugated and serum was separated after storage for one hour at room temperature. Biochemical analyses were done during the same day. Serum samples for measurement of leptin levels were stored at -20°C until they were used. Serum leptin levels were measured as ng/ml via immunoradiometric assay (IRMA) method by using Human Leptin IRMA DSL-23100 (Diagnostic Systems Laboratories, Inc. Texas, USA) kit. Following test procedures, test tubes were assessed with Gammabyt-CR gamma counter for one hour. Measurements for standards, controls and serums were repeated for confirmation. Sensitivity of the test was 0.10 ng/ml. Statistical analysis -------------------- Data were presented as median and range. Qualitative variables were assessed by Chi-square test. Between whole and sub-group comparisons were performed by non-parametric Kruskal-Wallis and Mann-Whitney U tests. A linear logistic regression analysis was performed with serum leptin levels as dependent variable and age, gender, BFM, aetiology of cirrhosis, Child-Pugh classification as independent variables in cirrhotics. A p value of \<0.05 was considered statistically significant. Results ======= Patient profiles and body composition ------------------------------------- Clinical and demographic characteristics of all and gender-based sub-groups were shown in table [1](#T1){ref-type="table"}. In male and female subjects, no statistically significant difference was observed in age, BMI, BFP and BFM between the controls and cirrhotic group (both, p \> 0.05). Following Child Pugh Classification, gender based or not, there were no significant differences in terms of BMI, BFP and BFM between controls and cirrhotic patients in each group according to the Child-Pugh classification (Figure [1](#F1){ref-type="fig"}) for each sex (Figure [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Characteristics of cirrhotic patients and controls in whole and gender based sub-groups. ::: Age Year BMI Kg/m^2^ BFP % BFM Kg Leptin ng/ml -------------------- ------------- ------------- ----------------- ---------------- ---------------------- Cirrhotic (n = 35) 53 (28--73) 24 (18--33) 27.9 (18.5--39) 19.4 (9.6--34) 13.5 (1.6--41)\* Female (n = 18) 48 (28--73) 24 (18--33) 32 (24--39) 20.1 (9.6--34) 15.5 (7.4--41)\\ Male (n = 17) 54 (35--66) 23 (19--27) 24.6 (18.5--28) 18.6 (11--26) 10.9 (1.6--36)\\\* Control (n = 15) 47 (37--65) 24 (20--26) 27.7 (22.4--37) 19.4 (14--26) 6.4 (0.14--16.3) Female (n = 7) 43 (37--61) 24 (22--25) 33.2 (30--37) 20 (18--26) 7.2 (5.58--16.3) Male (n = 8) 53 (42--65) 24 (20--26) 24.9 (22.4--28) 18.6 (14--21) 3.7 (0.14--8.7) Note: Data were presented as median and range. Groups and subgroups did not differ in terms of age, BMI, BFP, and BFM (p \> 0.05). \Cirrhotic vs. controls (p = 0.001), \\* cirrhotic females vs. control females (p = 0.025), \\\Cirrhotic males vs. control males (p = 0.002) BMI; Body mass index, BFP; Body fat percentage, BFM; Body fat mass ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Following Child Pugh Classification, there were no significant differences in terms of body mass index (BMI), body fat percentage (BFP) and body fat mass (BFM) between controls and cirrhotic patients (both, p \> 0.05). ::: ![](1471-230X-4-23-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Following Child-Pugh Classification, there were no significant differences in terms of body mass index (BMI), body fat percentage (BFP) and body fat mass (BFM) between female controls and female cirrhotic patients (both, p \> 0.05). ::: ![](1471-230X-4-23-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Following Child-Pugh Classification, there were no significant differences in terms of body mass index (BMI), body fat percentage (BFP) and body fat mass (BFM) between male controls and male cirrhotic patients (both, p \> 0.05). ::: ![](1471-230X-4-23-3) ::: Leptin levels ------------- Serum leptin levels were significantly higher in cirrhotic group than controls (p = 0.001) (Table [1](#T1){ref-type="table"}). There was a significant difference between the leptin levels of men and women in both control and cirrhotic groups (p = 0.029, p = 0.02, respectively) (Table [1](#T1){ref-type="table"}). Leptin levels were elevated in both female and male cirrhotics compared to controls (p = 0.025, p = 0.002, respectively) (Table [1](#T1){ref-type="table"}). Cirrhotic patients in each of A, B and C subgroups according to the Child- Pugh classification revealed significantly different leptin levels \[(ng/ml with median and range; 9.46 (1.6--30), 12.8 (4.2--18.8), 14.7 (8--41), respectively)\] compared to controls (ng/ml with median and range; 6.4 (0.14--16.3) (p = 0.046, p = 0.004, p = 0.0001, respectively). Gender based serum leptin levels of controls and cirrhotic cases that were grouped according to Child Pugh Classification were as shown in Figure [4](#F4){ref-type="fig"}. Male patients in the control group had significantly lower serum leptin levels compared to cirrhotic male cases that belongs to B and C classes (p = 0.006, p = 0.008, respectively). However, the difference was not significant between the control males and Child Pugh class A males (p = 0.234). On the other hand, female gender revealed significant difference only between Child Pugh C class patients and controls (p = 0.022). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Leptin levels in controls and cirrhotic patients by gender and Child-Pugh class. Male patients in the control group had significantly lower leptin levels compared to cirrhotic male cases that belongs to B and C classes (p = 0.006, p = 0.008, respectively). On the other hand, female gender revealed significant difference only between Child Pugh C class patients and controls (p = 0.02). In controls and Child Pugh B class patients, females had higher leptin levels than males. \P\< 0.02 vs. controls, in the same gender. ^◆^P\< 0.05 vs. different gender in the same group. ::: ![](1471-230X-4-23-4) ::: When age, gender, BFM, hepatitis B and C virus as etiologic factors of cirrhosis and child A, B and C as Child-Pugh classification were tested as independent variables for determination of serum leptin levels as dependent variable by linear logistic regression analysis in cirrhotic group, analysis result showed that Child-Pugh classification was the sole discriminator in determination of serum leptin levels in cirrhotic cases (beta: 0.435, p = 0.015) (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Linear regression analysis (R^2^= 0.326) with serum leptin as dependent variable in the cirrhotic group (n = 35). ::: Independent variables Beta p ----------------------------------- -------- --------- Gender (M-F) -0.307 0.065 Age (years) -0.227 0.183 BFM (kg) 0.006 0.974 Viral Etiologic Factor (HBV-HCV) 0.167 0.315 Child-Pugh Classification (A-B-C) 0.435 0.015\* Beta, beta regression coefficient; M, Male; F, Female; BFM, Body fat mass; HBV, Hepatitis B Virus; HCV, Hepatitis C Virus; A, Child-Pugh Class A; B, Child-Pugh Class B; C, Child-Pugh Class C. ::: Discussion ========== Leptin regulates body weight by suppressing appetite and increasing energy expenditure \[[@B1],[@B3],[@B4]\]. Anorexia and increased energy expenditure usually accompanies to the cirrhosis \[[@B17]\]. McCullough et al. reported modestly elevated circulating leptin levels in patients with alcoholic cirrhosis and they suggested that elevated serum leptin levels in cirrhosis might be responsible for the high prevalence of malnutrition among cirrhotic patients \[[@B11]\]. In our study, we also observed that circulating leptin levels were increased in non-alcoholic cirrhosis caused by viral hepatitis without severe energy malnutrition state. Leptin levels are higher in woman than in men \[[@B3],[@B6]\]. McCullough et al. found higher leptin levels among female cirrhotics than male cirrhotics, although the difference was not statistically significant \[[@B11]\]. These concepts are especially important in cirrhosis, because cirrhotics have gender-dependent alterations in body composition and sex steroids \[[@B18],[@B19]\]. When we considered gender in our study, serum leptin levels were significantly higher among females than males in both controls and cirrhotics. In addition, cirrhotic females and males had higher levels of serum leptin than the controls with the same gender. Since BMI and BFM values do not differ according to the sex and the presence or absence of cirrhosis, increased serum leptin levels could not be simply dedicated to BFM or malnutrition status in cirrhosis. In addition, linear regression test in the present study has shown that disease severity, which was determined by Child-Pugh classification, was the sole significant determinant of serum leptin levels in cirrhosis. In previous studies, association between the severity of cirrhosis and serum leptin levels is controversial \[[@B11]-[@B13]\]. Henriksen et al suggested that the elevated circulating leptin in patients with alcoholic cirrhosis was most likely caused by a combination of decreased renal extraction and increased release from subcutaneous abdominal, femoral, gluteal, retroperitoneal, pelvic, and upper limb fat tissue areas \[[@B20]\]. For this reason, we excluded cases with renal clearance impairment to avoid of accumulation of leptin in serum. In addition, using 4 different site of skinfold thickness measurement to calculate BFP and excluding cases with ascite that do not respond to diuretic therapy, we targeted to determine the relationship between body composition and serum leptin levels in controls and cirrhotics. In this study, BFM was found to associate with serum leptin levels in controls. However, BFM does not associate with serum leptin levels among cirrhotic patients. Therefore, we conclude that leptin production may differ among healthy and cirrhotic subjects. In an animal study, it has been shown that chronic ethanol consumption leads to increased serum concentrations of tumor necrosis factor and related cytokines such as leptin by inducing over production of these factors in the liver and peripheral adipose tissues \[[@B21]\]. Leptin secretion from adipocytes may be enhanced by cytokines released as a part of the inflammatory or fibrogenic process. Alternatively, as suggested, cirrhotic patients may simply exhibit decreased hepatic clearance of this protein \[[@B22]\]. Conclusion ========== Serum leptin levels increase in advanced liver disease independently of gender, body composition and viral etiologic factor in post-hepatitis cirrhosis. The increase is more abundant among patients that belong to C subgroup according to the Child- Pugh classification. Abbreviations ============= BMI, body mass index; BFP, body fat percentage; BFM, body fat mass Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= Bolukbas FF conceived of the study, and participated in its design and coordination. Bolukbas FF, Bolukbas C, Erdogan M and Zeyrek F collected the samples and carried out the laboratory analysis. Bolukbas C conceived of the study and participated in the sequence alignment and drafted the manuscript. Horoz M participated in the design of the study, participated in the sequence alignment and drafted the manuscript. Gumus M collected the clinical data and performed the statistical analysis. Yayla A drafted the manuscript and revised it critically for important intellectual content. Ovunc O participated in study design and coordination and revised the manuscript critically for important intellectual content. 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/4/23/prepub>	PubMed Central	2024-06-05T03:55:48.432943	2004-9-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522814/", "journal": "BMC Gastroenterol. 2004 Sep 23; 4:23", "authors": [ { "first": "Fusun F", "last": "Bolukbas" }, { "first": "Cengiz", "last": "Bolukbas" }, { "first": "Mehmet", "last": "Horoz" }, { "first": "Mahmut", "last": "Gumus" }, { "first": "Mehmet", "last": "Erdogan" }, { "first": "Fadile", "last": "Zeyrek" }, { "first": "Ali", "last": "Yayla" }, { "first": "Oya", "last": "Ovunc" } ] }
PMC522815	Background ========== In addition to the determination of immunoglobulin M (IgM) antibodies, the avidity of immunoglobulin G (IgG) antibodies is an important parameter for the diagnosis or exclusion of acute infections. Testing of IgG avidity has been applied for a large variety of pathogens (reviewed in \[[@B1],[@B2]\]). A correct diagnosis is especially important for infections during pregnancy with rubella virus, cytomegalovirus and Toxoplasma gondii. Therefore, avidity testing has been particularly useful for these pathogens. Determination of antibody avidity is usually based on the separation of low and high avidity antibodies by denaturing agents in enzyme immunoassays (EIA) or immunofluorescence assays. Several agents such as guanidine hydrochloride \[[@B3]\], diethylamine \[[@B4]\], thiocyanate \[[@B5]\] or urea \[[@B6]\] have been used for this purpose. These protein denaturants have been either included in the sample diluent (diluting principle) or in the washing buffer after the serum incubation step (eluting principle). Calculation of the avidity result has been performed in numerous ways. Mostly, avidity is expressed as percent ratio of antibody titers or EIA absorbance values with and without denaturation. Avidity results based on end-point titration with and without denaturant are considered to be the gold standard for avidity determination \[[@B7]\]. This technique has an excellent sensitivity and specificity and is considered not to be influenced by the concentration of the specific IgG. To reduce the considerable expense and labour required for the titration curves, simplified avidity tests based on single-point determinations using EIAs have been described \[[@B6],[@B8]-[@B11]\]. In these assays, EIA absorbance values or antibody titers mathematically derived from single dilutions have been used for avidity calculation. The avidity results based on single-point absorbance values are to some extent influenced by the concentration of specific IgG, but this is usually not critical for the distinction between low and high avidity. Diagnosis or exclusion of acute infection is the most common but not the only application of avidity determination. Determination of antibody avidity in pairs of serum and cerebrospinal fluid (CSF) samples has been suggested as a diagnostic means to detect intrathecal antibody synthesis \[[@B12]\] and to differentiate viral encephalitis from multiple sclerosis \[[@B13]\]. Longitudinal measurements of HIV avidity have been proposed to be useful for the assessment of HIV progression \[[@B14]\]. Because both of these applications involve comparative measurements of avidity where small differences may be of diagnostic importance, a technique for accurate avidity determination is required which is independent of antibody concentrations. The aim of our study was to establish an avidity assay that can be used for comparison of the avidities of varicella-zoster virus (VZV) IgG in serum and CSF. In order to achieve this aim, several modifications of a commercial VZV IgG assay were introduced. These modifications include the use of a novel denaturing agent and a new method to calculate the avidity results. Methods ======= Serum and CSF samples --------------------- The serum and CSF samples used in this study had been sent to the virological laboratory at the University of Würzburg for routine VZV testing and were stored in aliquots at -20°C. Two groups of serum samples were analyzed for the evaluation of the standard and modified VZV avidity assay. Group 1 consisted of 28 samples from 19 patients with acute or recent VZV infection and included 5 follow-up samples of patient F19, spanning a period of 11 months. The cases of this group fulfilled the following criteria: the presence of clinical symptoms suggestive of varicella, information on the disease onset, and a positive VZV IgM (Enzygnost Anti-VZV/IgM, Dade Behring, Marburg, Germany). Group 2 consisted of 37 samples from 37 subjects with infection in the distant past. In the subjects of this group, VZV IgG antibodies had been detected in previous samples taken at least 8 months earlier. For initial evaluation of the VZV avidity assay for CSF samples, three sample pairs of serum and CSF were tested, two from patients with VZV encephalitis (V2.1 and V9.2) and one from a patient with multiple sclerosis (M22). Standard VZV IgG avidity determination -------------------------------------- VZV IgG avidity determination of serum samples was performed in a semi-automated fashion using the Enzygnost Anti-VZV/IgG test kit (Dade Behring) with some modifications. Each of the sera diluted 1:231 in sample buffer was placed in two antigen coated microtiter plate wells. After 60 min at 37°C, one well was washed according to the instructions with the supplied washing buffer. The other well was washed with a solution of urea, or urea hydrogen peroxide, in the supplied washing buffer (for details, see Results) and once with the washing buffer only. The subsequent steps were carried out according to the instructions in an automated fashion using the Behring Elisa Processor III (BEP III; Dade Behring). Peroxidase-conjugated anti-human IgG was added in a 1:50 dilution and the plate was incubated at 37°C for 60 min. After three further washing steps with the supplied washing buffer, tetramethylene benzidine dihydrochloride was added as substrate and kept at room temperature for 30 min. The reaction was stopped with 0.5 N sulfuric acid. Control samples of acute and past VZV infections were included in each run. Where appropriate, VZV IgG antibodies were quantified by an one-point-quantification method (α-method, Dade Behring) according to the instructions of the manufacturer. VZV IgG avidity determination of CSF-serum-pairs ------------------------------------------------ To increase the sensitivity of the VZV IgG EIA in order to detect low-titer VZV IgG antibodies in CSF, the standard serum assay was modified as follows. The sample incubation time was increased to 180 min; the anti-human IgG was used in a 1:30 dilution and its incubation time was increased to 90 min. All incubations for the CSF assay were performed at room temperature because variation was found to be lower compared to incubation at 37°C. For the serum samples tested for evaluation purposes, the standard dilution of 1:231 was increased by a factor of 6 to yield a final dilution of 1:1386. Determination of VZV IgG avidity in CSF was always done in parallel with serum samples from the same time-point. Both serum and CSF were tested in at least four dilutions of a two-fold titration series. The starting dilution for each sample was derived from routine determinations of VZV IgG titers. For CSF samples, the starting dilution was at least 1:6. Calculation of the avidity index -------------------------------- VZV IgG avidity was calculated by various methods. First, it was calculated as the ratio of EIA absorbance values obtained with and without the denaturing washing step. Alternatively, one-point quantification titers (see above) instead of absorbance values were used for avidity calculation. Two additional methods were used for determination of VZV IgG avidity in CSF and the corresponding serum samples. One method involved the use of the software \"Avidity 1.2\", based on curve-fitting analysis of serial dilutions \[[@B15]\]. Secondly, we developed a new calculation method based on our observation from dilution series that the relationship between the absorbance values obtained without and with denaturing agents is linear. Thus, the relationship can be described by the equation y = m × x + c, where m is the slope, c the intercept, x the absorbance without the denaturant (abs~ref~) and y the absorbance with the denaturant (abs~denat~). The m- and c-values of the equation for each sample were calculated with the software Excel (Microsoft) from two or more experimentally derived data points. It is essential, that the absorbance values obtained without denaturant (abs~ref~) fall in the linear range of the enzyme immunoassay. For the VZV IgG assay with modified assay conditions, the linear range of the absorbance values without denturant extended from approximately 0.200 to 2.800. Avidity was determined from the linear equation for various virtual x-values as the percent ratio of the absorbance values with and without denaturing agent, i. e. avidity = (abs~denat~/abs~ref~) = (m × x + c)/x. The result was referred to as virtual absorbance ratio. Results ======= In order to evaluate the usefulness of the Enzygnost VZV IgG EIA for avidity testing, a panel of serum samples from patients with acute VZV infection of defined onset (group 1) and from controls with VZV infection in the distant past (group 2) was tested. In a preliminary experiment with three samples from each group, two different denaturing conditions were employed. The separation of the two sample groups with one 3 min washing step using 4 M urea hydrogen peroxide was superior to three 5 min washing steps with 5 M urea (Table [1](#T1){ref-type="table"}, Figure [1](#F1){ref-type="fig"}). Therefore, urea hydrogen peroxide was used for the following experiments. The results of avidity testing with urea hydrogen peroxide of all serum samples of group 1 and 2 are shown in Figure [2a](#F2){ref-type="fig"}. When avidity indices were calculated as the percent ratio of absorbance values, there was a clear distinction of the avidity values from both groups. Using a cut-off of 37 %, all samples from patients with disease onset of less than 50 days had avidity values below the cut-off, while all samples from the control group with VZV infection in the distant past had avidity values above the cut-off. Thus, a cut-off of 37 % resulted in a sensitivity and specificity of 100 % for the detection and exclusion of VZV infections within in the last 50 days. The standard result format of the Enzygnost VZV IgG assay are titers that are mathematically derived from single-point absorbance values (one-point-quantification). Because it had been shown previously for the Epstein-Barr virus (EBV) assay that avidity index calculations based on one-point-quantification titers gave results better than calculations using absorbance values \[[@B11]\], the avidity indices were recalculated using VZV IgG one-point-quantification titers instead of absorbance values. There was an excellent correlation between both methods of avidity index calculation, but in contrast to the EBV assay, the one-point-quantification method was not superior to that using absorbance values (data not shown). In general, IgG concentrations in CSF are by a factor of 1:200 to 1:1000 lower than in the corresponding serum. Therefore, the detection sensitivity of the VZV IgG assay was increased by prolonging incubations times and increasing the concentration of the anti-human IgG conjugate in order to reliably detect VZV IgG in all CSF samples from patients with positive serum VZV IgG. To study the effect of these assay modifications on avidity determination, all serum samples of group 1 and 17 randomly chosen samples of group 2 were retested with the modified assay. All the serum samples were tested in dilution 1:1386 instead of 1:231 used in the standard serum assay. The results are shown in Figure [2b](#F2){ref-type="fig"}. An avidity cut-off of 23 % resulted in a sensitivity of 95 % and specificity of 95 % for the detection and exclusion of VZV infections within the last 50 days, respectively. Although the standard serum conditions gave better results, there was again a clear separation between samples from both groups. Based on these results, the avidity assay with urea hydrogen peroxide and modified assay conditions to increase the sensitivity for VZV IgG detection was considered appropriate for use in VZV IgG avidity studies with CSF and is henceforth called the CSF avidity assay. For meaningful interpretation of avidity values in CSF, it is necessary to test serum samples obtained in parallel with the CSF. Because of the large difference of the IgG concentrations between serum and CSF, comparison of the avidity results necessitates independence of IgG concentration. To study this issue, CSF and serum samples were tested in twofold dilution series with the CSF avidity assay. Avidity was calculated as percent ratio of absorbance values from each dilution. The results for representative CSF and serum samples are shown in Figure [3](#F3){ref-type="fig"}. The avidity indices of most of the samples varied considerably depending on the sample dilution. For some samples, the range of avidity indices was greater than 20 %. The coefficient of variation (CV) was mostly higher than 10 % (examples in Table [2](#T2){ref-type="table"}). Thus, this method could not be used for comparative testing of serum and CSF samples. Alternatively, avidity indices were calculated with the software \"Avidity 1.2\", which is based on a mathematical model described previously by Korhonen et al. \[[@B15]\]. The software calculates avidity values from two sample dilutions, each tested with and without protein denaturant. To evaluate the influence of the working dilutions on the avidity indices obtained with this software, the avidity indices were calculated from different pairs of dilutions of the same sample. The results for representative examples are shown in Table [3](#T3){ref-type="table"}. The CV ranged from 5.9 % to 19.7 %. Because the two methods for avidity calculation described above were significantly influenced by the chosen working dilutions, we developed a new calculation method of antibody avidity. It is based on our observation that the relationship between absorbance values without and with denaturing agent is nearly linear. This relationship can therefore be described by the linear equation y = m × x + c, where m is the slope, c the intercept, x the absorbance without denaturant and y the absorbance with denaturant. Representative examples from patients with acute and past VZV infection and of serum and CSF pairs are shown in Figure [4](#F4){ref-type="fig"}. After experimental determination of the sample-specific m- and c-values, this observation allows calculating the y-value for any virtual absorbance without denaturation (x-value). Absorbance results from two dilutions are required in order to calculate the m- and c-values for a sample. In order to evaluate the influence of the chosen sample dilution on the avidity index derived from the virtual absorbance ratio, several samples were tested in series of twofold dilutions with four or more steps. The m- and c-values were calculated from different pairs of dilution steps for each sample. As a result of preliminary experiments, the absorbance 2.0 was chosen as the virtual x-value for the calculation of avidity values with the new method. (The linear range of the absorbances in the CSF assay format extended up to 2.8.) Table [3](#T3){ref-type="table"} shows that the CV for the avidity indices calculated by the new method from different pairs of dilution steps was lower than the CV obtained with the software \"Avidity 1.2\" from the same dilution series. Based on theoretical considerations we speculated that the slope of the linear equation by itself might represent the avidity of a sample, i.e. the steeper the linear curve, the higher the avidity. This was in agreement with the data from follow-up samples of a patient with acute VZV infection over a period of 11 months (Figure [5](#F5){ref-type="fig"}). However, comparison of the CVs demonstrated that avidity calculations using a fixed x were less influenced by the chosen dilution than avidity indices based on the slope (data not shown). Further dilution series of CSF and corresponding serum were tested with the CSF avidity assay in order to confirm that the influence of the IgG concentration on the avidity indices derived from the virtual absorbance ratio was minimal. Table [2](#T2){ref-type="table"} shows that the CVs with this method were equally low with CSF and serum. Results obtained with the software \"Avidity 1.2\" and results based on the absorbance ratios of single dilutions are presented for comparison. Although for two samples the CVs with \"Avidity 1.2\" were slightly lower than with the new method (CSF of V2.1 and CSF of V9.2), for the others the new method gave lower CVs. Avidity values derived from absorbance ratios resulted in the highest CVs for all samples. Discussion ========== We established a VZV avidity assay that is suitable for comparative evaluation of antibody avidity in serum and CSF samples. In order to achieve this aim several modifications of a serum VZV avidity assay were necessary. Though there have been a few studies on the use of avidity assays for VZV IgG antibodies \[[@B16]-[@B18]\], no standard method or commercial assays exist for the determination of VZV IgG avidity. Therefore, we first established conditions for a VZV IgG avidity assay that can be used for the differentiation of acute and past VZV infection. The serum assay was highly sensitive and specific for the diagnosis and exclusion of primary VZV infections. However, diagnostic applications of VZV IgG avidity determinations with serum samples are rare, because the diagnosis of acute or reactivated VZV infection is usually made clinically or by virus detection methods. Urea hydrogen peroxide appeared to be superior to urea in the optimization experiments for the VZV avidity assay and presents a novel option for a denaturing agent in avidity measurements. The denaturing effect of one washing step with a 4 M solution of urea hydrogen peroxide was more pronounced than that of three washing steps with a 5 M urea solution. Thus, the denaturing potential of urea hydrogen peroxide appears to be greater than that of urea. Furthermore, because of better solubility, handling of solutions of urea hydrogen peroxide is easier than handling solutions of urea in high molarity. Nevertheless, the optimal denaturing agent may vary between different antigens necessitating careful evaluation of denaturing conditions for each avidity assay. Under normal conditions, virus specific IgG is present in the CSF in very low concentrations. With an intact blood-CSF-barrier, the concentration gradient between serum and CSF is 200:1 to 1000:1. Thus, the usual serological methods for antibody quantification in serum are not sensitive enough to routinely detect specific IgG in CSF. Therefore, we modified the VZV IgG EIA to increase its sensitivity. This modification made it possible to detect and quantify VZV IgG in CSF samples from virtually all patients with measurable VZV IgG in serum. Corresponding serum samples were always run in parallel with CSF in the same assay. The CSF and serum dilutions were chosen individually for each sample pair in order to achieve similar absorbance values in the modified EIA. This assay formed the basis for avidity determination of CSF and the corresponding serum. For meaningful comparison of avidity values in serum and CSF, it is necessary to apply a calculation method that is independent of the concentration of specific IgG in the samples examined. One-point determination represents the simplest method for avidity index calculation, but such methods did not yield satisfactory results in this respect. The avidity technique based on end-point-titration is not influenced by the IgG concentration and is considered the reference method for avidity determination \[[@B1],[@B7]\]. However, it is relatively laborious and reagent consuming, requiring two distinct dilution series for the assays with and without protein denaturant. Therefore, we attempted to use the logistic approach based on end-point-titration, with fewer dilutions per sample \[[@B15]\]. This method works well in distinction between acute and past infections from samples of serum. However, under the conditions of the CSF avidity assay we found that the results of the logistic model were not uniformly independent of the sample working dilutions, albeit without straightforward dose-dependence of specific IgG. The reason for this limitation with our assay may lie in its modifications to increase sensitivity. Possibly, the curve fitting obtained with the logistic model would have to be adjusted in order to account for the special conditions of the CSF avidity assay. Searching further for a simple calculation method of avidity indices independent of IgG concentration and sample dilutions, we observed a linear relationship between the absorbance values of the assays without and with denaturing washing conditions. After performing an avidity assay of two dilutions of a given sample, this relationship can be exploited to calculate a linear equation from the two pairs of absorbance values (with and without denaturation). The only requirement is that all absorbance values should fall in the linear range of the EIA. Once for a given sample the linear equation is experimentally determined, the equation can be used for calculation of the virtual absorbance value with denaturation (abs~denat~) for any virtual absorbance value without denaturation (abs~ref~). The avidity index is then calculated as the ratio of the two virtual absorbance values (abs~denat~/abs~ref~). Thus, if a fixed abs~ref~is chosen for all samples to be compared, the avidity index becomes independent of the concentration of specific IgG in the samples. These theoretical considerations have been confirmed by the results obtained from dilution series in this study. It will be interesting to test the general applicability of this method for standard avidity assays by comparison with single-point and end-point avidity determinations. Conclusions =========== In summary, we have described several modifications of existing avidity techniques that have the potential to broaden the use of avidity assays in diagnosis and research. Urea hydrogen peroxide is a novel denaturing agent, which appears to be advantageous compared to urea in terms of handling conditions. The new calculation method of avidity indices based on a linear equation with fixed reference absorbance values is cost-effective and simple and has the potential to substitute other avidity calculation procedures. The CSF avidity assay in combination with the new calculation method allows for determination of the avidity of VZV IgG in corresponding serum and CSF samples with high precision and independent of the VZV IgG concentration. These features are necessary in order to study the avidity maturation in serum and CSF in patients with intrathecal synthesis of VZV-specific IgG antibodies. Competing interests =================== BW has received research and travel grants from Dade Behring. KH is employed by an organization (The Helsinki University Hospital Laboratory, HUSLAB) using the Avidity 1.2 software in infectious-disease diagnosis, and is a shareholder of an SME (Headman Ltd.) with commercial interest in it. Authors\' contributions ======================= RHK, JS and FT carried out the immunoassays and participated in the design of the study. RHK carried out the avidity index calculations. WZ participated in establishing the standard VZV IgG avidity assay. KH provided the software \"Avidity 1.2\" and participated in the data analysis. BW conceived of and coordinated the study and drafted the manuscript. All authors contributed to the final editing of the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2334/4/33/prepub> Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Optimization of the denaturing washing conditions. VZV IgG avidity of a panel of serum samples from patients with acute (▲) and past (■) VZV infection was tested with urea hydrogen peroxide (H~2~O~2~) and urea as denaturing agent. ::: ![](1471-2334-4-33-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Evaluation of different VZV IgG avidity assays with serum samples from patients with acute (▲) and past (■) VZV infection. The Enzygnost Anti-VZV/IgG test kit was used with standard conjugate concentration and standard incubation times (A) and with modified assay conditions (B; see text). ::: ![](1471-2334-4-33-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Influence of the concentration of specific IgG on single-point avidity indices. Two-fold dilution series of serum and CSF samples were tested in the CSF avidity assay and the avidity index was calculated as the percent ratios of absorbance values for each dilution. F19: Follow-up samples of a patient with acute VZV infection. V2.1 and V9.2: Pairs of serum (S) and CSF (C) of a patient with VZV encephalitis. ::: ![](1471-2334-4-33-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Linear relationship between absorbance values without and with denaturing washing step. Absorption-absorption-diagrams of dilution series of samples from (A) patients with acute (closed symbols) and past VZV infection (open symbols) and (B) of pairs of serum (closed symbols) and CSF samples (open symbols). For two examples, the determination of y-values for the virtual x-value 2.0 is represented graphically (dotted lines with arrow heads). Avidity is calculated as the ratio of absorbance y divided by absorbance x. ::: ![](1471-2334-4-33-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Absorption-absorption-diagram of follow-up serum samples of patient (F19) with acute VZV infection tested in the CSF VZV IgG avidity assay. F19.1/2/3/5: 15/28/42/336 days after onset of disease, respectively. ::: ![](1471-2334-4-33-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Comparison of the absorbance values with different denaturing washing conditions in the VZV IgG avidity assay. ::: ---------------------- -------------------- ----------------------------- --------- ----------- ------------- --------- VZV infection status 3 × 5 min 5 M urea 1 × 3 min 4 M urea H~2~O~2~ Abs (ref)^1^ Abs (denat)^2^ Avidity Abs (ref) Abs (denat) Avidity acute 2.221 1.208 54.4% 2.359 0.546 23.2% acute 1.829 0.676 37.0% 1.666 0.300 18.0% acute 1.982 0.457 23.1% 1.869 0.251 13.4% past 1.487 1.387 93.3% 1.448 0.850 58.7% past 1.190 0.830 69.7% 1.072 0.580 54.1% past 2.020 1.671 82.7% 1.929 0.984 51.0% ---------------------- -------------------- ----------------------------- --------- ----------- ------------- --------- ^1^absorbance value in EIA without denaturing washing step ^2^absorbance value in EIA with denaturing washing step ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of avidity values from serum and CSF pairs obtained with different calculation methods from different dilutions (absorbance ratios) or different pairs of dilutions (new method and \"Avidity 1.2\"). For each method, mean values, standard deviations (SD) and coefficients of variation (CV) are presented. ::: ----------------------- ------- ------ ------- ------ ------- ------ V2.1 V9.2 M22 Serum CSF Serum CSF Serum CSF absorbance ratios Mean (%) 50.2 54.0 59.9 66.4 34.7 38.6 SD (%) 5.6 5.4 7.8 8.5 4.1 6.1 CV (%) 11.1 10.0 13.0 12.8 12.0 15.9 Avidity 1.2 Mean (%) 37.6 45.4 50.5 56.5 23.2 31.9 SD (%) 3.7 3.0 3.0 4.6 2.3 3.6 CV (%) 9.9 6.6 5.9 8.1 9.9 11.4 new method Mean (%) 49.8 58.4 65.2 72.6 36.5 45.1 SD (%) 2.4 4.1 1.4 7.5 2.0 0.9 CV (%) 4.9 7.1 2.1 10.3 5.6 1.9 ----------------------- ------- ------ ------- ------ ------- ------ ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Calculation of avidity indices from dilution series of representative samples with the software \"Avidity 1.2\" and with the new calculation method described in this paper. ::: \(a) Raw data of dilution series -------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------- ---------------- ----------------- ---------------- ----------------- ---------------- Serum F19.3 Serum P28 Serum V9.2 Relative dilution^1^ Abs (ref)^2^ Abs (denat)^3^ Abs (ref) Abs (denat) Abs (ref) Abs (denat) 1:1 2.765 0.925 2.668 1.600 2.936 1.948 1:2 1.962 0.626 1.833 1.144 1.943 1.268 1:4 1.428 0.356 1.300 0.727 1.269 0.782 1:8 0.831 0.190 0.824 0.441 0.778 0.463 1:16 0.592 0.106 0.480 0.252 0.409 0.192 \(b) Avidity indices obtained from different combinations of dilutions with arithmetic mean, standard deviation (SD) and coefficient of variation (CV) Serum F19.3 Serum P28 Serum V9.2 Dilutions used for calculation Avidity 1.2 (%) new method (%) Avidity 1.2 (%) new method (%) Avidity 1.2 (%) new method (%) 1:1 & 1:2 9.5 32.0 33.4 61.8 46.1 65.4 1:2 & 1:4 15.2 32.3 40.4 63.7 49.2 65.5 1:4 & 1:8 18.0 25.8 41.5 57.4 50.0 62.9 1:8 & 1:16 11.7 30.0 44.6 54.4 56.6 68.0 1:1 & 1:4 11.6 30.0 35.5 58.7 48.2 64.7 1:2 & 1:8 17.3 32.0 41.8 63.0 50.1 65.4 1:4 & 1:16 13.5 26.4 43.5 56.6 50.2 64.2 1:1 & 1:8 14.7 31.7 41.3 59.0 52.8 65.2 1:2 & 1:16 15.0 32.0 42.0 62.7 51.3 65.4 Mean 14.1 30.2 40.4 59.7 50.5 65.2 SD 2.8 2.5 3.7 3.3 3.0 1.4 CV 19.7 8.4 9.0 5.5 5.9 2.1 ^1^Relative dilution: the starting dilutions were 1:1512 (F19.3), 1:144 (P28), 1:399168 (V9.2) ^2^absorbance value in EIA without denaturing washing step ^3^absorbance value in EIA with denaturing washing step :::	PubMed Central	2024-06-05T03:55:48.434264	2004-9-8	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522815/", "journal": "BMC Infect Dis. 2004 Sep 8; 4:33", "authors": [ { "first": "Ralf-Herbert", "last": "Kneitz" }, { "first": "Jörg", "last": "Schubert" }, { "first": "Franz", "last": "Tollmann" }, { "first": "Wolfgang", "last": "Zens" }, { "first": "Klaus", "last": "Hedman" }, { "first": "Benedikt", "last": "Weissbrich" } ] }
PMC522816	Background ========== Norovirus(NV, previously \"Norwalk-like viruses\"), one of four genera in the Caliciviridaefamily, includes a group of morphologically similar but genetically different single-stranded RNA viruses. NVs represent the most important cause of non-bacterial gastroenteritis worldwide. In industrialised countries NVs may be responsible for up to 80% of all outbreaks of gastroenteritis \[[@B1]\]. Outbreaks may affect all age groups and generally occur in crowded communities such as restaurants, tourist resorts, hospitals, schools and nursing homes. Contaminated food or water commonly represents the main source of infection. Epidemics spread by the faecal-oral route, even if transmission may also occur through direct person-to-person contact or aerosolised viral particles. The incubation period of NV gastroenteritis is 24--48 hours and symptoms include vomiting, diarrhoea, abdominal pain, low-grade fever, headache and myalgia. Italy has no specific surveillance system for viral gastroenteritis and laboratory diagnosis is only carried out in a few cases. Therefore the impact of NV infection is currently unknown and little information is available about circulating strains. Outbreak investigations are usually performed by local health units. Gastroenteritis notifications are often too delayed to identify the etiologic agent and source of infection correctly. Puglia, a region in the South-East of Italy, has about four million inhabitants. Bari (about 1,500,000 people in the province) is the capital. Raw mussels are largely consumed in Bari and, in general, in Puglia, especially at Christmas and Easter. In 1994 raw seafood consumption was the source of a small cholera outbreak. Moreover, hepatitis A is endemic in this region. This report describes a large outbreak of NV gastroenteritis that involved several households in Bari province. Methods ======= Descriptive and analytical studies ---------------------------------- The outbreak occurred during the Easter holidays, between March 31^st^and April 7^th^2002, in the province of Bari (Puglia region, South Italy). In Italy, at Easter many people eat in restaurants and the day after Easter most people go picnicking. In the province of Bari seafood (especially mussels) are often consumed on these occasions. The Regional Epidemiological Office, alerted by the Local Health Unit, performed a field study, collecting data from all General Practitioners and all Emergency Units in the province. Active surveillance was conducted until April 15^th^2002 (one week after the last case onset). The aim of the investigation was to describe the outbreak and to identify the etiologic agent, the source of infection and the means of transmission. A case was defined as illness in a resident in the area during the period March 31^st^-- April 7^th^, who had diarrhoea (three or more loose stools in any 24 hour period) or vomiting (at least one episode). Fever greater than or equal to 38°C, abdominal pain or nausea were considered additional symptoms. A \"probable\" secondary case was defined as illness in a household with onset of symptoms more than 24 hours after the primary case. A retrospective cohort study was performed in order to assess risk factors associated with illness. All households where a case occurred were included in the study. Information, collected by means of a standardised questionnaire, included: 1) demographic individual characteristics such as age, gender, occupation; 2) type of food consumed during the last 72 hours before onset of symptoms; 3) if ill, type, date and time of onset of symptoms. To assess the association between food consumption and disease, relative risks (RR) and 95% confidence intervals (95% CI) were calculated. Age and gender were compared between ill and unaffected individuals by the chi squared and Mann-Whitney tests. A p-value less than 0.05 was considered significant. Variables associated to the illness in the univariate analysis were included in the stratified analysis; summary RR and 95% CI were calculated from the formulas of Greenland and Robins \[[@B2]\]. Collected data were analysed by using Epi Info 6.04 (Centres for Disease Control and Prevention, Atlanta, GA) and Statview 4.0 (Sas Institute Inc., Cary, NC) software. Laboratory investigations ------------------------- Faecal specimens were collected from ill individuals. Part of the specimens was refrigerated and processed within 12 hours to detect ova and parasites by direct microscopy and common bacteria by standard methods. The rest was stored at -20°C until examination by NV-specific reverse transcription-polymerase chain reaction (RT-PCR). Viral nucleic acid extraction and purification from stool specimens was performed as previously reported \[[@B3]\]. RT-PCR was carried out with the primers JV12-SM31 specific for the polymerase gene of NV \[[@B4]\]. Eleven samples of mussels were collected from two fish-markets from which the cases had bought the shellfish they consumed. Mussel samples were also processed within 12 hours to detect common bacteria by standard methods. Then a nested PCR was used. The procedure for mussel processing as well as viral RNA extraction and purification has been previously described in full \[[@B5]\]. The primers used for first round PCR were JV12 and SM31. Nested PCR was carried out with the use of the primers SR33 for negative strand DNA synthesis and SR48 and SR46 for positive strand synthesis of genogroup I (GI) and genogroup II (GII) sequences, respectively \[[@B6]\]. The 333-bp or 123-bp amplification products from cases and from mussels were subjected to sequencing with PCR primers. When required, cloning was carried out on PCR products. Sequences obtained were aligned with those available in the GenBank. Results ======= Epidemiological issues ---------------------- One hundred and three subjects met the case definition; 22 of them were defined as \"probable\" secondary cases. Fifty eight (56.3%) of the 103 cases were female; the mean age was 42.6 years (table [1](#T1){ref-type="table"}). The clinical pattern of the disease was characterised by the presence of vomiting (84.5%), nausea (58.3%), diarrhoea (53.4%), abdominal pain (47.6%), fever (16.5%). No difference was observed in the clinical pattern by gender. Age was significantly higher in ill individuals with severe diarrhoea (lasting more than 24 hours or undergoing specific treatment) (46.3 vs 38.3 years; p = 0.036). No difference in age was found for the other symptoms. The outbreak started on March 31^st^at 5,00 pm and lasted eight days. The epidemic curve shows a single peak on April 2^nd^(35 cases between 1:00 and 12:00 AM) and a right tail probably due to secondary cases (figure [1](#F1){ref-type="fig"}). Incubation time was not calculated because it was not possible to state a single exposure time. In fact this outbreak could represent the result of multiple small household outbreaks, the source of which could have been either a restaurant meal (on Easter Sunday) or a picnic the day after Easter. All cases belonged to 30 households that included 139 individuals in all. All members of each household ate together at least once (at the same restaurant or during the same picnic) in the days before the onset of symptoms of relatives. All 36 healthy subjects were interviewed. Nineteen (52.8%) of them were female, the mean age was 42.5 years. Gender and age were not significantly different between healthy and ill individuals (table [1](#T1){ref-type="table"}). An attack rate of 74.1% (103/139) was observed. In the univariate analysis, the association between raw mussel consumption and illness was significant (RR: 1.50; 95% CI: 1.18 -- 1.89; p \< 0.001). The attack rate according to raw mussel consumption was 86.3% (69/80). Even cooked mussel consumption was associated with illness in the univariate analysis (RR: 1.53; 95% CI: 1.05--2.23; p = 0.003). No other food was associated with illness (table [2](#T2){ref-type="table"}). Stratified analysis showed that cooked mussel consumption was significantly associated with illness only among those who did not eat raw seafood (RR: 3.04; 95% CI: 1.26 -- 7.30; p \< 0.001). Such association was not significant in the \"ate raw mussels\" stratum (RR: 0.84; 95% CI: 0.76 -- 0.93; p \> 0.05). The relative risk according to cooked mussel consumption was 1.38 (95% CI: 1.00 -- 1.91) after correction for raw mussels consumption (table [3](#T3){ref-type="table"}). Subjects defined \"probable secondary cases\" were considered ill according to the risk assessment. In fact, we could not be certain that \"probable secondary cases\" were actually co-primary cases; on the other hand, the association between raw mussels and illness was stronger excluding from analysis such cases (RR: 1.63; 95% CI: 1.22--2.18; p \< 0.001). Laboratory issues ----------------- Both stool and mussel samples were negative for parasites and bacterial enteropathogens. All stool samples from 24 cases were positive for NV by RT-PCR and 5 of 11 mussel samples by nested PCR. The sequences of strains revealed great heterogeneity. In fact, the simultaneous occurrence of GI and GII viruses within the same outbreak was observed. One sample of mussels showed the presence of a mixed genogroup. Sequence analysis showed that strains from 19 cases and 3 mussel samples had identical sequences and belonged to GII, clustering narrowly to the strain NLV/Tarrag/238/2001/Sp, thus strengthening the epidemiologic link of the mussels to the cases. The nucleotide sequences from three cases and 2 mussel samples formed two other distinct clusters showing the best fit with the strains Saitama U25 and Khs1-1997-JP. Sequences from two further cases and one mussel sample were assigned to genogroup I and showed a high degree of identity with the strain NLV/Steinbach/EG/2001/CA. Three familial groups consisting of 7 cases and three members of a further family group from the outbreak showed identical sequences and were classified as NV GII (the main cluster of 19 cases). A fourth member of the latter family group was positive for one of the two GI strains. The other strains investigated by sequence analysis were unrelated to those from these familial groups. Discussion ========== In Italy a national NV outbreaks database does not exist, although NVs are the leading cause of gastroenteritis outbreaks in Europe as in the rest of the western countries \[[@B7]\]. Moreover, to our knowledge, this is the first NV community outbreak to be confirmed in Italy \[[@B8]\]. The actual number of NV gastroenteritis cases is underestimated since the illness is often mild and diagnostics are provided only in a few laboratories. Despite these factors, the knowledge of circulating strains and the recognition of the common vehicles of infection are of primary importance for prevention of such disease. This investigation showed the causative role of mussel consumption in the outbreak, confirmed by laboratory tests on both stools and food samples. The heterogeneity of viral strains associated with the consumption of contaminated mussels is no surprise due to the peculiar features of shellfish. In fact, bivalve shellfish are filter feeders and tend to accumulate whatever pollutants are in the water which can result in viral contamination from a multitude of possible sources affecting many individuals. In any case, further studies should be carried out to clarify this issue \[[@B9]\]. The low relative risks we found could be due to the study design. In fact, the selection of households where a case occurred could underestimate the association. However this study design became necessary because of the unfeasibility of an open cohort study. In the 1990s, NVs were identified as the primary pathogen associated with shellfish-borne gastroenteritis in the United States \[[@B10]\]. Since then many studies have confirmed the role of shellfish in the spread of NV infection. Raw shellfish consumption is very frequent among the people of Puglia. Mussels sold in Puglia come from a large number of suppliers: from countries within the European Union (EU) and from other Mediterranean countries. The frequent shellfish consumption had already been blamed for the cases of cholera that occurred in Puglia in 1994, during the Christmas period, when a major epidemic was reported in Albania, on the opposite coast of the Adriatic sea \[[@B11]\]; moreover, mussels have been identified as the principal vehicle during recurring hepatitis A epidemics \[[@B5],[@B12],[@B13]\]. A significant presence of HAV in mussels sold in Puglia has recently been demonstrated by experimental evidence \[[@B3],[@B5]\], even in samples negative for standard microbiological controls. In Puglia informative campaigns and routine microbiological controls on seafood are currently performed. The effectiveness of these interventions seems to be scarce. Moreover, our data showed cooked mussels also played a significant role. In fact cooking might not completely inactivate NVs and because of the low infectious dose, even a limited residual contamination can result in illness \[[@B14]\]. The epidemic curve shows a usual pattern consistent with a common point source. However, person to person transmission may have played a role in the last phase of the outbreak. On the other hand, such a means of transmission of NV is well documented. This investigation confirms the importance of field study in gastroenteritis outbreaks, as well as obtaining stool specimens from patients and food samples for laboratory analysis in a short time \[[@B15]\]. The present standards for human faecal contamination do not seem to be a reliable indicator of viral contaminants in mussels \[[@B5]\]. Seafood samples analysed during this investigation were all negative for the presence of common bacteria. On the other hand, to protect consumers it would be necessary to use a molecular index of the human contamination. In such cases reference laboratories with high-technology facilities would be required. The lack of such laboratories could be an obstacle to implementing a routine control system based on molecular tests. Conclusions =========== This episode confirms that large NV outbreaks occur in Italy but only an accurate investigation can recognise this pathogen and that current regulations and commercial practices need to be revised to assure the safety of shellfish consumption and to improve control of future outbreaks. Competing interests =================== None declared. Authors\' contributions ======================= MC and GB carried out microbiological assays and drafted the manuscript. CG participated in the design of the study and performed the statistical analysis. MQ conceived the study as well as participating in its design and coordination. 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/1471-2334/4/37/prepub> Acknowledgements ================ This study was conducted within Italian Ministry of University Research Project (MIUR -- PRIN 2003). Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Distribution of cases of NV gastroenteritis (n= 103) by onset time, March-April 2002. Probable secondary cases are coloured in grey. ::: ![](1471-2334-4-37-1) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Study population by age and gender. ::: Ill Not Ill p --------------------------------- ------------- ------------- ------------ Age (mean ± standard deviation) 42.6 ± 20.1 42.5 ± 20,8 \>0.05\* Age range 5 -- 80 6 -- 85 Gender (M/F) 45/58 17/19 \>0.05\\ Total 103 36 \* unpaired t test \\ chi square ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Attack rates, relative risk, confidence intervals and chi-squared p value according to exposure. ::: Exposure No. Exposed No. Cases (n = 103) Attack Rate (%) Relative Risk 95% Confidence Interval p value --------------------- ------------- --------------------- ----------------- --------------- ------------------------- --------- --------- raw mussels 80 69 86.3 1.50 1.18 1.89 \<0.001 cooked mussels 112 89 79.5 1.53 1.05 2.23 0.003 eggs 16 13 81.3 1.10 0.85 1.43 \>0.05 oven cooked pasta 62 43 69.4 0.89 0.73 1.09 \>0.05 grilled sausage 51 36 70.6 0.93 0.75 1.15 \>0.05 grilled meat 89 65 73.0 0.96 0.79 1.17 \>0.05 grilled lamb 70 52 74.3 1.01 0.83 1.22 \>0.05 dairy milk products 77 54 70.1 0.89 0.73 1.08 \>0.05 cheese 49 35 71.4 0.93 0.76 1.15 \>0.05 fresh vegetables 85 65 76.5 1.09 0.88 1.34 \>0.05 pastries 53 40 75.5 1.03 0.84 1.26 \>0.05 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Relative risk and 95% confidence interval according to mussels consumption: stratified analysis. ::: No. Exposed No. Cases Attack Rate (%) Relative Risk 95% Confidence Interval p value -------------------------------------------------------------- ------------- ----------- ----------------- --------------- ------------------------- --------- --------- eating raw mussels = \"yes\" (n = 80)eating cooked mussels 70 59 84.3 0.84 0.76 0.93 \>0.05 eating raw mussels = \"no\" (n = 59)eating cooked mussels 42 30 71.4 3.04 1.26 7.30 \<0.001 Summary Risk Ratio 1.38 1.00 1.91 :::	PubMed Central	2024-06-05T03:55:48.437899	2004-9-21	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522816/", "journal": "BMC Infect Dis. 2004 Sep 21; 4:37", "authors": [ { "first": "Rosa", "last": "Prato" }, { "first": "Pier Luigi", "last": "Lopalco" }, { "first": "Maria", "last": "Chironna" }, { "first": "Giovanna", "last": "Barbuti" }, { "first": "Cinzia", "last": "Germinario" }, { "first": "Michele", "last": "Quarto" } ] }
PMC522817	Background ========== Carcinoma of the esophagus is an aggressive malignancy with an increasing incidence. Its virulence, in terms of symptoms and mortality, justifies a continued search for optimal therapy. The large and growing number of patients affected, the high mortality rates, the worldwide geographic variation in practice, and the large body of good quality research warrants a clinical practice guideline. This clinical practice guideline was developed by the Gastrointestinal Cancer Disease Site Group (DSG) of Cancer Care Ontario\'s Program in Evidence-based Care (PEBC), using the methods of the Practice Guidelines Development Cycle \[[@B1]\]. This practice guideline report is a convenient and up-to-date source of the best available evidence on neoadjuvant or adjuvant therapy for resectable esophageal cancer, developed through systematic review, evidence synthesis, and input from practitioners in Ontario. The PEBC has a formal standardized process to ensure the currency of each clinical practice guideline report. This process consists of the periodic review and evaluation of the scientific literature and, where appropriate, integration of this literature with the original clinical practice guideline information. The systematic review on neoadjuvant or adjuvant therapy for resectable esophageal cancer, which forms the basis for this clinical practice guideline, is available in a companion document \[[@B2]\]. Based on the systematic review, draft recommendations were developed by consensus of the Gastrointestinal Cancer DSG to create the clinical practice guideline report. The clinical practice guideline is intended to promote evidence-based practice in Ontario, Canada. As part of the PEBC\'s clinical Practice Guideline Development Cycle, all draft recommendations are sent to Ontario practitioners for external review. The efficacy of this external review process has been previously described \[[@B3]\]. The external review is a mailed survey consisting of items that address the quality of the draft practice guideline report and draft recommendations and whether the draft recommendations should serve as a practice guideline. Final approval of this practice guideline report was obtained from the Practice Guidelines Coordinating Committee (PGCC). Methods ======= Clinical practice guideline development --------------------------------------- ### Systematic review A systematic review with meta-analysis on neoadjuvant or adjuvant therapy for resectable esophageal cancer was developed by the Gastrointestinal Cancer DSG of Cancer Care Ontario\'s Program in Evidence-based Care \[[@B2]\]. The evidence examined did not support the use of neoadjuvant or adjuvant chemotherapy or radiotherapy for resectable thoracic esophageal cancer. Gastrointestinal cancer disease site group consensus ---------------------------------------------------- In discussions regarding the completed systematic review, the Gastrointestinal Cancer DSG agreed that the evidence did not support a recommendation for neoadjuvant or adjuvant chemotherapy or radiotherapy for resectable thoracic esophageal cancer. A recommendation that surgery alone should be the standard of care for this patient population was drafted, and it was recommended that the draft practice guideline be sent out to Ontario practitioners for external review. The role of radiotherapy alone and chemoradiation alone without surgery is addressed in a separate Gastrointestinal Cancer DSG Clinical Practice Guideline: Combined modality radiotherapy and chemotherapy in the non-surgical management of localized carcinoma of the esophagus\[[@B4]\]. Results ======= External review --------------- Practitioner feedback was obtained through a mailed survey of 163 practitioners in Ontario (27 medical oncologists, 21 radiation oncologists, 112 surgeons, and three gastroenterologists). The survey consisted of items evaluating the methods, results, and interpretative summary used to inform the draft recommendations and whether the draft recommendations should be approved as a practice guideline. Written comments were invited. Follow-up reminders were sent at two weeks (post card) and four weeks (complete package mailed again). The Gastrointestinal Cancer DSG reviewed the results of the survey. Eighty-six surveys (58%) were returned. Twenty-nine respondents (34%) (nine medical oncologists, seven radiation oncologists, and 13 surgeons) indicated that the report was relevant to their clinical practice and completed the survey. Key results of the practitioner feedback survey are summarized below. 1\. Number surveyed: 163 practitioners in Ontario, Canada involved in the care of cancer patients 2\. Return rate: 58% (mean Gastrointestinal Cancer DSG return rate: 60.2%; range: 51% -- 84%) 3\. Written comments attached: 10% 4\. Agreement with the summary of evidence: 90% 5\. Agreement with the recommendation: 90% 6\. Approval of the recommendation as a practice guideline: 69% ### Summary of main findings Three (10%) respondents provided written comments. One practitioner hypothesized that preoperative chemoradiation might have a role in adenocarcinoma of the lower third of the esophagus (as suggested by Walsh et al \[[@B5]\] with 100% adenocarcinoma and by Urba et al \[[@B6]\] with 75% adenocarcinoma), but not in squamous cell carcinoma (as suggested by Bosset et al \[[@B7]\] and by Le Prise et al \[[@B8]\]). Another respondent noted that the survival advantage at three years for combined treatment for preoperative chemoradiotherapy is discounted in the guideline report, and suggested that the guideline recommend the selection of the option preferred by informed patients. There was a request for an algorithm to help in deciding between surgical and non-surgical treatment. The same respondent commented on the limited discussion on quality of life. Two radiation oncologists disagreed with the recommendations and thought that the draft practice guideline report should not be approved as a practice guideline, but neither provided written comments. Discussion ========== Gastrointestinal cancer disease site group modifications and actions -------------------------------------------------------------------- After completion of the practitioner feedback survey, additional trials were found. The results of two randomized trials both found surgery alone to be significantly superior to radiation alone \[[@B9],[@B10]\], which resulted in an original draft recommendation regarding radiation alone as a primary modality for localized esophageal cancer being removed from the final practice guideline. In response to this feedback, the Gastrointestinal Cancer DSG acknowledged that the majority of studies have been performed in squamous cell carcinomas. While adenocarcinomas were included in some studies, a distinction between the two histological subtypes was not made because previous studies have not consistently found that they respond differently to chemotherapy or radiation, and nine references \[[@B11]-[@B19]\] were added to support this. The Gastrointestinal Cancer DSG did not feel the evidence was compelling enough to recommend preoperative chemoradiotherapy over surgery alone based on the three-year data. After consideration, the Gastrointestinal Cancer DSG decided not to create an algorithm as suggested as a similar project is currently under development. After addressing the comments obtained from practitioners during the external review, the Gastrointestinal Cancer DSG voted that the overall guideline recommendations should be approved, and submitted the practice guideline to the Practice Guidelines Coordinating Committee for review. Practice guidelines coordinating committee approval process ----------------------------------------------------------- The practice guideline report was circulated to members of the Practice Guidelines Coordinating Committee for review and approval. All 11 members of the PGCC returned ballots. Eight PGCC members approved the practice guideline report as written and three members approved the guideline conditional on the Gastrointestinal Cancer DSG addressing specific concerns. PGCC members requested that the following issues be addressed prior to the approval of the guideline report: One member noted that although the majority of studies had been performed in squamous cell carcinomas, some studies included adenocarcinomas, and it would be helpful if the pathological subtypes were discussed. In particular, this member wanted to know if there was any difference in response or outcome for the two histological subtypes. Another member noted that although the pooled analysis for preoperative chemoradiation versus surgery alone detected no difference at one year, the pooled estimate almost reached significance. This member was concerned that the discussion may be too dismissive of the data, and suggested there be some acknowledgment that further follow-up and additional studies are needed. In response to this feedback, the Gastrointestinal Cancer DSG expanded on the earlier revisions concerning the similarities in response to treatment between squamous cell carcinomas and adenocarcinomas. Also, after the original practice guideline was submitted to the PGCC, two meta-analyses \[[@B20],[@B21]\] both detecting a statistically significant difference in survival at three years favouring preoperative chemoradiation versus surgery alone were obtained. The Gastrointestinal Cancer DSG re-pooled the mortality data from the six trials \[[@B5]-[@B8],[@B22],[@B23]\] at three years and obtained similar results. Conclusions =========== In consideration of the systematic review, external review, and subsequent Practice Guidelines Coordinating Committee revision suggestions, and final approval, the Gastrointestinal Cancer Disease Site Group developed the following Clinical Practice Guideline: Practice guideline ------------------ This practice guideline reflects the most current information reviewed by the Gastrointestinal Cancer DSG. ### Target population These recommendations apply to adult patients with resectable and potentially curable thoracic (lower two-thirds of esophagus) esophageal cancer for whom surgery is considered appropriate. ### Recommendation • If surgery is considered appropriate, then surgery alone (i.e., without neoadjuvant or adjuvant therapy) is recommended as the standard practice for resectable thoracic esophageal cancer. This Clinical Practice Guideline report is based on work completed in October, 2003. All approved PEBC Clinical Practice Guideline reports are updated regularly. Please see the PEBC\'s web site <http://www.cancercare.on.ca/access_PEBC.htm> for a complete list of current and on-going projects. Competing interests =================== The author(s) declare that they have no competing interests. List of abbreviations used ========================== In order of appearance: DSG, Disease Site Group; PEBC, Program in Evidence-based Care; PGCC, Practice Guidelines Coordinating Committee. Authors\' contributions ======================= RM, RW, and LZ created the initial drafts of this clinical practice guideline with input from other members of the Gastrointestinal Cancer DSG. RM, RW, and BR created the final draft of this clinical practice guideline. Creation of the submitted manuscript was performed by BR and RM. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/4/67/prepub> Acknowledgements ================ Additional members of Cancer Care Ontario\'s Program in Evidence-based Care Practice Guidelines Initiative\'s Gastrointestinal Cancer Disease Site Group include: O. Agboola MD, M. Citron, F.G. DeNardi MD, S. Fine MD, B. Fisher MD, C. Germond MD, D. Jonker MD, K. Khoo MD, W. Kocha MD, M. Lethbridge, W. Lofters MD, and V. Tandan MD. Please see the Practice Guidelines Initiative (PGI) web site <http://www.cancercare.on.ca/access_PEBC.htm> for a complete list of current and past Disease Site Group members.	PubMed Central	2024-06-05T03:55:48.440891	2004-9-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522817/", "journal": "BMC Cancer. 2004 Sep 24; 4:67", "authors": [ { "first": "Richard A", "last": "Malthaner" }, { "first": "Rebecca KS", "last": "Wong" }, { "first": "R Bryan", "last": "Rumble" }, { "first": "Lisa", "last": "Zuraw" } ] }
PMC522818	Background ========== Undifferentiated nasopharyngeal carcinoma (NPC) is a malignant epithelial tumor characterized by a heavy infiltration of non malignant lymphocytes and most of these tumor infiltrating lymphocytes (TIL) have been shown to be T cells \[[@B1]\]. The Epstein-Barr virus (EBV) is causally associated with this malignancy since viral DNA is regularly present in the malignant epithelial cells but not in the neighbouring normal tissues. In addition, NPC patients show a specific pattern of humoral responses against EBV antigens \[[@B2]\]. Viral proteins known to be expressed in NPC tumor cells are the EBV-encoded nuclear antigen 1 (EBNA-1) and the latent membrane proteins LMP-1 in 35 to 65% of cases, and LMP-2 \[[@B3],[@B4]\]. The latent membrane proteins have been shown to serve as targets for EBV-specific cytotoxic T lymphocytes (CTL) from normal seropositive individuals \[[@B5],[@B6]\]. Recently, CD8 positive EBV-specific cytotoxic T cell clones were isolated from the peripheral blood and tumors of NPC patients \[[@B7]\]. The majority of the isolated CTL clones are directed towards the most immunogenic EBNA3 proteins which are not expressed in NPC tumor cells. No EBV-specific CTL activity is detectable by the standard chromium release assay in NPC patients \[[@B8]-[@B10]\] and the activity of any CTLs that would be present in such patients appears to be somehow suppressed. This lack of cytotoxic activity is in sharp contrast with the strong anti-EBV humoral immune response seen in patients \[[@B11],[@B12]\]. The discrepancy between these two types of immune responses in NPC is still unexplained. It has been hypothesized that some viral gene products might have the capacity to influence cytokine production in such a way as to inhibit specific CTL activity \[[@B3],[@B13]\]. Interestingly, the product of the EBV BCRF1 open reading frame has been found to display extensive homology with human interleukin 10 \[IL-10 ; \[[@B14]\]\]. Like its human counterpart, this viral product designated vIL-10, exerts immunosuppressive functions \[[@B15]\]. It is postulated that IL-10 production in malignant tumors may facilitate their escape from immune surveillance \[[@B16]\]. The expression of IL-10 in NPC has been controversial. While it has been reported that IL-10 is not expressed by NPC cells as detected by RNA in situ hybridisation \[[@B17]\], some reports using immunohistochemical and molecular techniques showed the expression of this cytokine by epithelial NPC tumor cells and TIL \[[@B18]-[@B20]\]. These authors suggested IL-10 as a possible evasion mechanism against the host antiviral system. Such a mechanism would explain the lack of detection of EBV specific cytotoxic activity in NPC patients at both peripheral and intratumoral levels \[[@B8]-[@B10],[@B21]\]. Indeed, IL-10 is known to inhibit cell-mediated immune responses \[[@B22]\]. IL-10 is also known for upregulating the B cell response \[[@B23]\] and therefore, this putative mechanism is in accordance with the strong EBV-specific humoral immune response seen in NPC \[[@B11],[@B12],[@B24]\]. Other interleukins such as IL-2 and IL-6 may also appear to be involved in this discrepancy between humoral and cellular immune responses due to their central regulatory effects on T or B cells \[[@B25],[@B26]\]. In this report, we investigated the ability of both peripheral blood lymphocytes (PBL) and TIL of undifferentiated NPC patients to express three interleukins (IL-2, IL-6, IL-10) and three immunoglobulin isotypes (IgM, IgG, IgA) following pokeweed mitogen (PWM) stimulation in vitro. The data obtained indicated some significant differences between NPC patients and controls in interleukin and immunoglobulin production. However, further investigations are needed to establish the relevance of these differences to the discrepancy between humoral and cellular immune responses characteristic of NPC. Methods ======= Patients and controls --------------------- 17 untreated Tunisian NPC patients were included in this study. Informed consent was obtained from all patients before collection of blood and biopsy samples. These patients presented for treatment at the National Cancer Institute Salah Azaiez, Tunis or at the Farhat Hached hospital, Sousse, Tunisia. Their mean age was 47,5 years (range 25--70). All patients were diagnosed histopathologically as undifferentiated NPC. 17 healthy EBV carriers who presented to donate blood at the National Blood Transfusion Centre, Tunis, Tunisia, were included as controls in this study. Sera of all patients and controls were titrated for antibodies directed against EBV antigens by indirect immunofluorescence \[[@B27]\]. Lymphocyte preparations ----------------------- Peripheral blood and biopsy samples were taken on the same day from each of the 17 NPC patients. Peripheral blood was collected from patients and healthy donors in heparin. Mononuclear cells were isolated by centrifugation over Ficoll-Hypaque (Pharmacia) according to Boyum \[[@B28]\]. Following three washes in RPMI-1640 medium (Gibco), cells were resuspended in RPMI-1640 medium supplemented with 10% heat-inactivated fetal calf serum (FCS ; Gibco) and 50 μg/ml Gentamicin (complete medium) at a concentration of 2 × 10^6^cells/ml. For the preparation of tumor-infiltrating lymphocytes, biopsies were collected from NPC tumors under sterile conditions and immediately transferred to RPMI-1640 medium supplemented with 50 μg/ml Gentamicin. The biopsies were washed several times in this medium and then minced into small pieces in order to extract lymphocytes. The extracted cells were then washed by centrifugation in the same medium and then resuspended in complete medium at a final concentration of 2 × 10^6^cells/ml. Samples from each lymphocyte preparation were taken for immunophenotyping by indirect membrane immunofluorescence using mouse monoclonal antibodies directed against CD3, CD4, CD8 or CD19 and FITC-conjugated goat anti-mouse F(ab\')~2~fragment. The labelled cells were counted under a fluorescence microscope. Stimulation by Pokeweed mitogen ------------------------------- PBL from healthy donors or NPC patients and TIL from the same patients suspended at 2 × 10^6^cells/ml in complete medium were distributed at 1 ml / well in 24-well Costar plates. One hundred microliters of a predetermined optimal dilution of PWM (Gibco), for treated wells, or complete medium alone, for untreated controls, were added in corresponding wells. The cells were then cultured for 12 days at 37°C, 5% CO~2~and 98% relative humidity in a CO~2~incubator. At the end of this incubation period, culture supernatants were harvested and used for cytokine and immunoglobulin determination as described below. Cytokine and Immunoglobulin determination ----------------------------------------- Culture supernatants of PBL and TIL stimulated by PWM under optimal conditions were used for both cytokine and immunoglobulin determination by enzyme-linked immunosorbent assays (ELISA). IL-2, IL-6 and IL-10 concentrations were measured using commercial sandwich-type ELISA kits (Immunotech enzyme immunoassays, France) according to the procedures described by the manufacturer. IgM, IgG and IgA concentrations were measured using sandwich-type ELISA assays prepared in our laboratory. Briefly, 96 well-microplates (Greiner, Germany) were coated for 2 hours at 37°C and overnight at 4°C with 150 μl/well of isotype-specific mouse monoclonal antibody (Sigma, France) appropriately diluted in 0,1 M carbonate buffer, pH 9.6. The microplates were washed four times with phosphate-buffered saline pH 7.4 containing 0,1 % Tween 20 (PBS-Tween), then blocked with 300 μl/well of PBS containing 1% FCS for 30 minutes at 37°C. For the assay, 100 μl /well of appropriately diluted culture supernatants in PBS-Tween containing 10% FCS were incubated in coated microplates for 2 hours at 37°C. The microplates were then washed four times with PBS-Tween and 100 μl/well of the appropriate alkaline phosphatase-labeled conjugate (Goat anti-human IgM, IgG or IgA, Sigma-France) were added at a 1:20000 dilution in PBS containing 1% FCS. After a two-hour incubation at 37°C, the microplates were washed and a p-nitrophenylphosphate substrate solution was added. The microplates were incubated for 1 hour at 37°C and the reaction was stopped with 50 μl/well of 1 N NaOH. Immunoglobulin standards were generated using purified human immunoglobulins of each isotype (Sigma, France). The microplates were read at 405 nm on a microplate reader (LP-400, Diagnostic Pasteur). Data were represented as the mean immunoglobulin concentration of triplicate cultures. Immunoglobulin values were obtained by interpolation from standard curves. Statistical analysis -------------------- Data were analysed using \"Student\'s t-test\". Probability values below 5% were considered significant. Results ======= EBV serology ------------ All patients showed a typical serological profile characteristic of NPC as determined by indirect immunofluorescence. Levels of IgG antibodies to VCA were considerably higher than those in healthy donors. Their titers varied from 640 to 2560 (mean titer = 1065), whereas in healthy donors such titers fluctuated between 40 and 160 (mean titer = 70). IgG antibodies to EA were present in NPC patients only, with titers ranging from 20 to 320 (mean titer = 50). Anti-VCA IgA antibodies were detected in all patients (range : 20--320 ; mean = 61) and anti-EA IgA antibodies were detected in only 4 patients (range : 10--80 ; mean = 27). None of the controls showed IgA antibodies against these antigens. Immunophenotyping of lymphocyte preparations -------------------------------------------- Immunophenotypic analysis of the lymphocyte preparations obtained from patients and controls indicated that CD3+ lymphocytes constituted the major subpopulation with a mean frequency of (60 ± 8)% in PBL of both patients and controls and (55 ± 9)% in TIL. CD4+ lymphocytes represented (35 ± 5)% in PBL of both patients and controls and (20 ± 3)% in TIL. CD8+ cells showed mean values of (40 ± 6)% in PBL of patients, (36 ± 4)% in TIL, and (28 ± 5)% in PBL of controls. On the other hand, CD19+ B lymphocytes represented in average (25 ± 5)% in PBL of patients, (23 ± 6)% in TIL, and (22 ± 3)% in PBL of controls. Cytokine production ------------------- As illustrated in Figures [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}, the PBL of NPC patients produced significantly more IL-2 (2105 ± 1152 pg/ml) and IL-10 (1280 ± 727 pg/ml) than the PBL of healthy donors (1286 ± 452 pg/ml, p = 0,022 for IL-2 and 793 ± 325 pg/ml, p = 0,016 for IL-10). Comparable amounts of IL-6 were found for both groups (2375 ± 919 pg/ml for patients and 2177 ± 435 pg/ml for controls, p = 0,429 ; Figure [3](#F3){ref-type="fig"}). On the other hand, TIL of patients showed a significantly higher IL-2 (3913 ± 1484 pg/ml, p = 0,0002) and IL-10 (1926 ± 817 pg/ml, p = 0,02) production than their PBL. The observed differences in mean IL-6 production between patients\'PBL (2375 ± 919 pg/ml) and TIL (1962 ± 515 pg/ml) did not reach statistical significance (p = 0,116). Unstimulated lymphocytes cultured in the absence of PWM did not show any detectable cytokine production (data not shown). Immunoglobulin production ------------------------- The results of immunoglobulin determination showed that the PBL of NPC patients produced significantly higher levels of IgM (8262 ± 5315 ng/ml) than the PBL of controls (3753 ± 2801 ng/ml, p = 0,004 ; Figure [4](#F4){ref-type="fig"}). Both groups produced similar amounts of IgG (874 ± 408 ng/ml for patients and 847 ± 442 ng/ml for controls, p = 0,85 ; Figure [5](#F5){ref-type="fig"}) and IgA (4380 ± 4316 ng/ml for patients and 3067 ± 2267 ng/ml for controls, p = 0,27 ; Figure [6](#F6){ref-type="fig"}). On the other hand, TIL of these patients produced significantly higher levels of IgM (21162 ± 12276 ng/ml, p = 0,0003) and IgG (2789 ± 1583 ng/ml, p \< 0,0001) than their PBL. No significant differences in IgA production were found between PBL (4380 ± 4316 ng/ml) and TIL (6112 ± 6046 ng/ml, p = 0,34). Unstimulated lymphocytes cultured in the absence of PWM did not show any detectable immunoglobulin production (data not shown). Discussion ========== Previous reports indicate the absence of EBV-specific cytotoxic T lymphocytes detectable by the standard chromium release assay in both the peripheral and intratumoral compartments \[[@B8]-[@B10],[@B21]\] in undifferentiated nasopharyngeal carcinoma. However, NPC patients show a strong EBV-specific humoral immune response, and elevated titers of anti-EBV antibodies directed against viral antigens are observed in their sera \[[@B11],[@B12],[@B24]\]. The mechanisms underlying this discrepancy between humoral and cellular immune responses in NPC patients are still unknown. It was postulated that interleukin 10 production in malignant tumors would facilitate their escape from immune surveillance \[[@B16]\]. The expression of IL-10 in NPC has been controversial. While some authors have reported that NPC tumor cells do not express IL-10 \[[@B17]\], others observed IL-10 expression in epithelial NPC tumor cells and tumor infiltrating lymphocytes \[[@B18],[@B19]\]. They suggested such IL-10 expression as a possible mechanism for NPC tumors and EBV to escape local cellular immune attack. Indeed, IL-10 is a pleiotropic factor known for its suppressive effects on cell-mediated immune responses \[[@B22]\]. In sharp contrast to these inhibitory effects, IL-10 also has a potent stimulatory effect on the humoral immune response, inducing B lymphocyte differentiation and immunoglobulin secretion \[[@B23]\]. Therefore, IL-10 hyperproduction alone or in association with changes in other cytokines might lead to an imbalance between humoral and cellular immune responses similar to that seen in NPC. The expression of cytokines such as IL-2 \[[@B19]\], IL-6 \[[@B17]\] and IL-10 \[[@B18],[@B19]\] in NPC has been studied mainly on tumor biopsies using immunohistochemical and molecular techniques. To the best of our knowledge, only one study on the ability of NPC patients\'lymphocytes to produce cytokines in culture has been reported and it was limited to IL-2 \[[@B9]\]. In this report, we investigated the ability of both peripheral blood and tumor infiltrating lymphocytes of 17 undifferentiated NPC patients to produce cytokines following mitogenic stimulation in culture. Since immunoglobulin isotypes are determined by cytokine patterns, we also looked for possible correlations between measured cytokine levels and immunoglobulin isotypes produced. PWM was chosen to stimulate the lymphocyte cultures because it is known to activate both T and B lymphocytes in humans \[[@B29],[@B30]\]. In addition, the ability of PWM to stimulate cytokine production by Tcells is similar to that of PHA \[[@B31]\]. This allowed us to study the responses of both T and B cells simultaneously in each lymphocyte culture. The data obtained indicate that the highest levels of IL-2 were produced by TIL cultures followed by patients\'PBL. This is in line with a report by Lakhdar et al \[[@B9]\] showing a higher IL-2 production by PHA-stimulated PBL of NPC patients than by controls. Such high IL-2 levels are expected to favor a strong cytotoxic response since IL-2 is needed for CTL stimulation and proliferation \[[@B32]\] and they are not consistent with the known lack of detectable CTL activity. IL-10, a representative of the Th2 pattern of cytokines, is generally considered immunosuppressive. It inhibits IL-6 secretion by activated macrophages but not by Th2 lymphocytes \[[@B33]\]. In the present work, IL-10 was overproduced by patients\'lymphocyte cultures whereas their IL-6 levels were similar to controls, showing no signs of inhibition by IL-10. This points to Th2 lymphocytes rather than activated macrophages as the main source of IL-6 in this system. The increased ability of patients lymphocytes to produce IL-10 is compatible with the lack of CTL activity. Cultures of tumor infiltrating lymphocytes secreted significantly higher amounts of IgM and IgG than the PBL of either patients or controls in good agreement with the increased levels of IL-2 and IL-10, since these cytokines are involved in B cell activation and enhance immunoglobulin synthesis \[[@B23],[@B25]\]. No significant correlations between cytokine production and IgA secretion were found, in line with the preferential enhancing effect of IL-2 and IL-10 on IgM and IgG synthesis \[[@B23]\]. In an attempt to see whether the observed differences in cytokine and immunoglobulin production between patients and controls or between PBL and TIL of patients could be due to differences in the composition of individual lymphocyte preparations, we performed an immunophenotypic analysis of each lymphocyte preparation. As shown in the results, the only significant changes in the proportions of lymphocyte subsets between patients and controls were observed in the CD8 subpopulation which showed an increase in patients PBL and TIL. On the other hand, the CD4 subset showed a significant decrease in TIL. Such differences in lymphocyte subpopulations are not expected to produce the changes in cytokine and immunoglobulin production observed here, since CD4 lymphocytes are well known for being the main source of IL-2 and IL-10, and for their ability to stimulate immunoglobulin synthesis \[[@B34]\]. Recently, CD4+ lymphocyte populations have been shown to be more heterogeneous and complex than previously thought \[[@B35]\] and new subsets of T regulatory cells (Tr) with immunosuppressive activities have been identified. Tr cells have been suggested to play a key role in the evasion from immune-mediated clearance of microorganisms and tumors \[[@B36]\]. It has also been reported that IL-10-producing Tr1 cells dominate the immune response to LMP1 in EBV seropositive subjects \[[@B36]\]. Since LMP1 is expressed in NPC tumors, it would be tempting to speculate that the increase in IL-10 production by TIL would correspond to an increase in Tr1 cells in the corresponding NPC tumors. In this respect, Hodgkin lymphoma infiltrating lymphocytes have been shown to contain large populations of both Tr1 and CD4+CD25+ regulatory T cells \[[@B37]\], and it would be interesting to see whether a similar situation occurs in NPC. Conclusion ========== In conclusion, our data point to the possibility of an imbalance in immunoregulatory interleukin production in NPC patients. Their lymphocytes, especially those infiltrating the tumors, showed in particular a high propensity to produce IL-10 following mitogenic stimulation in vitro. Overproduction of this pleiotropic cytokine may lead to a discrepancy between humoral and cellular immune responses similar to that seen in NPC. The relevance of the present observations to the in vivo situation is not yet established and warrants further investigations. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= LFJ designed and carried out the experimental work, performed the statistical analysis and drafted the manuscript. NK maintained the cell cultures and provided technical help. RHK and KB participated in the design and coordination of the study. RK directed the research and finalized 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/1471-2407/4/68/prepub> Acknowledgements ================ We thank Dr Youssef Gharbi for valuable suggestions and help during this work. We also thank the Ministère de la Santé Publique and the Secrétariat d\'Etat à la Recherche Scientifique, Tunisia, for funding. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### IL-2 production. PBL and TIL from 17 NPC patients and PBL from 17 controls were cultured at a concentration of 2 × 10^6^cells/ml in the presence of PWM at 1:200 dilution for 12 days. Culture supernatants were collected and assayed for IL-2. ::: ![](1471-2407-4-68-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### IL-10 production. PBL and TIL from 17 NPC patients and PBL from 17 controls were cultured at a concentration of 2 × 10^6^cells/ml in the presence of PWM at 1:200 dilution for 12 days. Culture supernatants were collected and assayed for IL-10. ::: ![](1471-2407-4-68-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### IL-6 production. PBL and TIL from 17 NPC patients and PBL from 17 controls were cultured at a concentration of 2 × 10^6^cells/ml in the presence of PWM at 1:200 dilution for 12 days. Culture supernatants were collected and assayed for IL-6. ::: ![](1471-2407-4-68-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### IgM production. PBL and TIL from 17 NPC patients and PBL from 17 controls were cultured at a concentration of 2 × 10^6^cells/ml in the presence of PWM at 1:200 dilution for 12 days. Culture supernatants were collected and assayed for IgM. ::: ![](1471-2407-4-68-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### IgG production. PBL and TIL from 17 NPC patients and PBL from 17 controls were cultured at a concentration of 2 × 10^6^cells/ml in the presence of PWM at 1:200 dilution for 12 days. Culture supernatants were collected and assayed for IgG. ::: ![](1471-2407-4-68-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### IgA production. PBL and TIL from 17 NPC patients and PBL from 17 controls were cultured at a concentration of 2 × 10^6^cells/ml in the presence of PWM at 1:200 dilution for 12 days. Culture supernatants were collected and assayed for IgA. ::: ![](1471-2407-4-68-6) :::	PubMed Central	2024-06-05T03:55:48.442114	2004-9-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522818/", "journal": "BMC Cancer. 2004 Sep 27; 4:68", "authors": [ { "first": "Lilia", "last": "Fliss-Jaber" }, { "first": "Radhia", "last": "Houissa-Kastally" }, { "first": "Kamel", "last": "Bouzouita" }, { "first": "Naceur", "last": "Khediri" }, { "first": "Ridha", "last": "Khelifa" } ] }
PMC522819	Background ========== In the rat the first 3 weeks of life constitute a critical period of neuromuscular plasticity. Contractile properties of muscles are not inherent, but are determined by the motor nerve, that supplies the muscle. This was shown by experiments of crossinnervation between fast and slow muscles \[[@B1]\]. Axotomy of peripheral nerves in neonatal rats, leads to loss of the bigger immature motoneurons through an excitotoxic process, with bigger neurons firing at a greater frequency. Consequently there is loss of the fast-contracting muscle fibers being innervated by bigger nerve cells \[[@B2]\]. There is much evidence that overactivation of glutamate receptors plays a significant role in this process (glutamate excitotoxicity) \[[@B3]\]. Glutamate is a major neurotransmitter in the CNS. Ionotropic receptors of glutamate (NMDA and AMPA / kainate) have been identified throughout the brain and spinal cord in several species of animals, including humans \[[@B4]\]. Their activation leads to Ca^2+^influx into the cell and subsequent activation of a cell death cascade (activation of proteases, lipases and other enzymes leading to cell lysis). Following neonatal sciatic nerve injury, the surviving motoneurons, take at least 8 days to grow back to the hind limb muscles, whereas most of the motoneurons that die, do so by apoptosis within the first two days \[[@B5]\]. However, it has been shown by previous studies \[[@B6],[@B7]\] that motoneurons are highly vulnerable to excitotoxic cell death, only during the first five days of postnatal life. Sciatic nerve injury by axotomy in neonatal rats has been shown to result in significant reduction in the number of surviving motoneurons in the ventral horn of the lumbar segments. Developing motoneurons become resistant to axon injury when their axon terminals are converted from growing into transmitting structures and form stable connections with the muscle fibers they innervate \[[@B8]\]. Target deprivation from muscle after the 5^th^postnatal day (P5) does not cause significant motoneuron loss, since damaged axons reinnervate muscle fibers successfully \[[@B6],[@B7],[@B9]\]. In this study we examine the possible neuroprotective effects of systemic administration of magnesium ions, after axotomy in neonatal rats on the 2^nd^postnatal day. It could be assumed that in our model, magnesium ions penetrate the blood brain barrier, as magnesium has been shown to concentrate in the cerebrospinal fluid after intraperitoneal injection and increase the electrical threshold required to control seizures, in a rat model, by other investigators \[[@B10]\]. We also investigated the time course of motor unit loss and the alterations in the contractile properties of a fast- (EDL) and slow-contracting (soleus) muscle. Methods ======= Experimental procedures ----------------------- Wistar Albino rats of both sexes were used in these experiments. All efforts were made in order to minimize the number of animals used and their suffering. The research project complies with the guidelines for animal use, established by the American Physiological Society and was approved by the local ethical committee in accordance with EEC Council Directive 86/609. The day of birth was taken as P0 (zero). At the 2^nd^postnatal day the sciatic nerve of the left hindlimb was crushed, in order to perform axotomy. Four experimental groups were included in this study: i) control, ii) rats whose sciatic nerve was crushed (axotomy), iii) magnesium treated controls and iv) rats whose sciatic nerve was crushed and were magnesium treated. The rats were examined for the contractile properties and the motor unit number of two hindlimb muscles (EDL and soleus) at several stages of postnatal development. Tension recordings were held in both hindlimbs (operated and control) in groups of Mg^2+^-treated and non-treated rats at: a) Postnatal day 14 (P14), b) Postnatal day 21 (P21), c) Postnatal day 28 (P28) and d) Adult rats (older than 2 months). In our study six successfully tested Mg^2+^-treated and six non-treated rats of each age-group (48 rats) were included. Nerve crush ----------- The animals were anesthetized with ether at the 2^nd^postnatal day and were operated under sterile conditions. Surgery was performed under an operating stereoscope (10× magnification). The sciatic nerve was identified and crushed at the mid-thigh, just proximal to its division to the tibial and common peroneal nerves. Care was taken not to damage the blood supply to the surrounding tissues. Crush was performed using a fine forceps for 30 seconds. Then the nerve was examined to ensure that the epineural sheath was intact but translucent (axotomy). The wound was closed with fine sutures. All surgical procedures were carried out by the same surgeon. Three hours after wound suturing the neonates were placed back with their mothers. Specific tests were performed daily in order to confirm the efficacy of axotomy: a\) The plantar- and the dorsi-flexion reflexes were checked: The examiner, placing his index finger on the animals foot, forces it to plantar- and dorsi-flex the ankle. The animal reflects actively, doing the opposite movement, if able to dorsi- and plantar-flex, respectively. No reactive movements are evident after successful denervation. b\) Animals were suspended by their tail and the inability of normal movement of the left hindlimb, indicating successful nerve axotomy, was assessed for the following \[[@B11]\]: \_ Hip and knee in extension \_ Ankle in the operated side in plantar flexion \_ Adduction of the whole limb \_ Weakness of digit extension in operated side The first behavioral signs of reinnervation after axotomy, should be evident at about 10--12 days after injury, according to other investigators \[[@B12]\]. Only those animals, whose successful axotomy was verified, were included in our study. Magnesium administration ------------------------ The rats of some litters were treated daily with magnesium ions. Mg^2+^was injected subcutaneously as a solution of MgSO~4~7H~2~O (0,05 ml of 1 M solution / 10 g body weight). This was the highest tolerable dose. Mortality in higher doses exceeded 80%. The treatment continued until the rats were 14 days old. Control groups were injected daily with the same volume of normal saline (0,05 ml / 10 g body weight). In Mg^2+^treated animals weight gain slowed down during the first month of their life. Treated and control adult rats, however, did not show statistically significant differences in their body weight. Axotomized rats, presented with motion deficit on the denervated limb, as described above, for about 2 weeks. After Mg^2+^treatment normal kinetic behaviour and reflexes returned on the 7^th^to 9^th^postnatal day. Tension recording ----------------- Animals were anesthetized with Chloral Hydrate (4,5% 1 ml / 100 g of body weight). The EDL and soleus muscles of both the axotomized and the contralateral control leg were prepared. The distal tendons were dissected free and attached to a strain gauge transducer (Dynamometer UFI, Devices) by a short silk suture and the exposed parts of the muscles were kept moist with warm (37°C) Krebs solution (NaCl 118,08 mM, NaHCO~3~25 mM, glucose 5,55 mM and CaCl~2~1,89 mM) \[[@B13]\]. The sciatic nerve was exposed. In order to isolate soleus muscle contraction and to avoid summation of concomitant gastrocnemius muscle contraction, after stimulating the sciatic nerve we used to cut the branches innervating the gastrocnemius and plantaris longus muscles. The leg was held rigidly in a position of 90° flexion of the knee and ankle joints, by two pins, inserted in the femoral condyles and the calcaneus, respectively. The tendons of EDL and soleus muscles were connected to a strain gauge and the tension elicited by sciatic nerve stimulation (Digitimer DS9A stimulator) was displayed on the screen of an oscilloscope (Fluke PM 3380 A). Muscle length was adjusted so as to produce maximal single twitch tension (optimal length), through a micromanipulator allowing motion on the 3 axes (Prior, England). Then stimulus intensity was adjusted in order to elicit maximal tension, using supramaximal (3--9 volts) square pulses of 0,5 msec duration. The signal from the transducer was amplified by a DC transducer amplifier (Neurolog NL 107). Time To Peak (TTP) and Half Relaxation Time (HRT) of the Single Twitch recording were measured. Tetanic contractions were then elicited by stimulating the nerve at 10, 20, 40, 80 and 100 Hz. The duration of stimulus was 250 msec. All devices during the tension-recording procedure were controlled by a pulse programmer (Digitimer D4030). The number of motor units (MU) was estimated by the incremental method (the number of different single twitch tensions produced by stepwise increments of stimulus intensity). The fatigueability of the muscles was tested by stimulating them at 40 Hz for 250 msec every second. The decrease in tension at 3 minutes of such stimulation was measured and the % Fatigue Index (FI) was calculated: F.I. = (Initial Tension - Tension at 3 min) / Initial Tension. After tension recordings were completed, muscles were excised and weighed. The animals were killed by cervical dislocation. Statistical analysis -------------------- Stiatical analysis of the results was performed using SPSS 10.0 software for Windows. Nonparametric tests (Mann -- Whitney for two independent variables and Kruskal -- Wallis for more than two independent variables) were used in order to compare data, of different groups. Criterion of statistical significance was set at p \< 0.05. Results and discussion ====================== Number of motor units --------------------- The number of motor units was estimated by the stepwise increments of tension, created by stimuli of different intensity. Our findings are in consistency with results of previous work by other authors \[[@B14]\], such as by our previous experience \[[@B7]\]: The control EDL muscle contains approximately 40 motor units, whereas soleus of 30 motor units, independently of animal age (Table [1](#T1){ref-type="table"}, Figures [1a](#F1){ref-type="fig"} and [1b](#F1){ref-type="fig"}). Treatment with Mg^2+^does not affect the number of motor units of control muscles. Axotomy at P2 results in statistically significant (p \< 0.05) motor unit loss in both EDL and soleus in all age groups. This means that motoneuron death is already established at P^14^. Treatment with Mg^2+^results in a statistically significant difference in the survival of motor units compared to non-treated axotomized rats. It is obvious that the neuroprotective effect of Mg^2+^is already established at P14 immediately after the period of treatment (Table [1](#T1){ref-type="table"},[2](#T2){ref-type="table"},[3](#T3){ref-type="table"} and Figures [1a](#F1){ref-type="fig"} and [1b](#F1){ref-type="fig"}). Figure [2](#F2){ref-type="fig"}, images the different single twitch tension recordings, elicite, in an EDL muscle on the left leg (side of nerve injury), after stimulation of the left sciatic nerve with electrical stimuli of incrementally increasing intensity. In this case, the axotomized EDL muscle consisted of six motor units, that survived axotomy. Tension development ------------------- Axotomy affects tension development by the muscle (Tables [4](#T4){ref-type="table"} and [5](#T5){ref-type="table"}). In adults rats single twitch of EDL is 4.63 ± 0.78% of the control muscle, whereas this of soleus is 16.80 ± 3.03% (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). Maximal Tetanic Tension is being developed by stimulation at 100 Hz. Maximal Tetanic Tension in adult rats is only 3.31 ± 0.30% of the control side, whereas after Mg^2+^treatment it is 15.16 ± 0.89% respectively. The excessive discrepancy of force outcome ability by both muscles between P28 and adulthood is also noticeable: At P28 respective values are: 18.94 ± 4.16 vs. 40.14 ± 19.34%. At P21 respective values are: 23.35 ± 16.03% vs. 64.22 ± 43.02% (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). The absolute values of Tetanic Tension at 100 Hz as listed in Tables [4](#T4){ref-type="table"} and [5](#T5){ref-type="table"}, are statistically significant different, between the operated (left) and the control (right) side. At P14, on the other hand, a tension development deficit does not develop after axotomy. Maximal Tetanic Tension of soleus is also being reduced after axotomy: 12.44 ± 0.97% of the control side in adult rats. Respective values in the other age groups are: 67.39 ± 39.21% at P28, 79 ± 14.34% at P21 and 81.96 ± 13.56% at P14 (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). This marked reduction of tension developing ability of both EDL and soleus, in adult rats, is established after the first month of life. An obvious explanation is the excessive muscle atrophy that occurs gradually as the animal grows up. Mg^2+^treatment, which results in inhibition of muscle atrophy, has been shown to promote motor units survival, and causes the axotomized soleus to remain as strong as the control muscle. Maximal Tetanic Tension of axotomized compared to control side is 97 ± 11.33% in adult rats, 91.45 ± 15.09% at P28, 77.01 ± 25.63% at P21 and 92.12 ± 11.15% at P14 (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). Figure [2](#F2){ref-type="fig"} shows representative recordings of twitch and titanic tensions, as well as fatigue indexes of EDL and soleus muscles; tables [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"} and [5](#T5){ref-type="table"} summarize all the above results. Contraction velocity -------------------- EDL is normally a fast contracting muscle in adult rats, whereas soleus is a slow one. Immature (P^14^) muscles, however, are not yet differentiated into fast- or slow-contracting. In adult rats TTP and HRT are respectively a) 36 ± 4.4 msec and 28 ± 2.28 msec in the EDL and b) 56.2 ± 3.37 msec and 59.5 ± 3.78 in the soleus (Table [6](#T6){ref-type="table"}). At P14 in the EDL, TTP is 56 ± 8.16 msec and HRT is 59 ± 4.34 msec, whereas the respective values for soleus are 74 ± 8.29 msec and 66 ± 8 msec. Axotomy gradually converts EDL into a slow-contracting muscle: TTP is 56 ± 4.97 at P14, 49 ± 10.18 at P21, 43 ± 5.48 at P28 and 77 ± 7.89 in adult rats (Table [6](#T6){ref-type="table"}). These values are significantly different to the ones obtained from control muscles. Soleus on the other hand, remains slowcontracting following axotomy, in all age groups. Axotomy does not alter soleus contraction velocity (no statistically significant difference between values). Mg^2+^administration did not affect contraction and relaxation velocity of control muscles (no statistically significant difference). It caused axotomized EDL to become fast-contracting in adult rats (TTP: 38 ± 7.53 and HRT: 43 ± 4.13), as it is predicted for control muscles, whereas soleus\' contractility was not affected. Fatigueability -------------- Soleus is a fatigue resistant muscle in adult rats, whereas EDL is not. However, both immature EDL and soleus muscles show properties of fatigueable muscles. EDL (control or operated) at P14 is not fatigue resistant, irrespective of Mg^2+^administration, with a fatigue index of about 65% (Table [6](#T6){ref-type="table"}). During normal development EDL remains fatigable to adulthood, with Mg^2+^treatment not affecting this state. Axotomy however, causes EDL to become fatigue resistant in adult rats (F.I = 15.6). Values of operated side compared to those of control muscles, are statistically significantly different even at P21. Mg^2+^administration after axotomy does not induce conversion of EDL to a fatigue resistant muscle. ### Soleus While soleus is not fatigue resistant at P14 (F.I.= 55.6%), it gradually becomes fatigue resistant during normal development (Table [6](#T6){ref-type="table"}). Axotomy on the other hand, results in soleus becoming less fatigue resistant in adult rats (F.I. = 34.7%, statistically significant different than control muscles). However, if axotomy is combined with Mg^2+^treatment, the development of soleus into a fatigue resistant muscle is not hindered. Muscle weight ------------- Muscle weight on the operated side is statistically significantly reduced compared to the contralateral control side. It is noticeable, however, that there is a marked reduction in muscle weight from P28 to adulthood both in the EDL and the soleus. The reduction in muscle weight of axotomized EDL was already established at P14, whereas in soleus there was a statistically significant difference only after P28 (Tables [4](#T4){ref-type="table"} and [5](#T5){ref-type="table"}). Mg^2+^administration provokes muscle weight increase on the operated side. In soleus, values are almost equal to those of control muscles, whereas in EDL the weight gain is less dramatic (Tables [4](#T4){ref-type="table"} and [5](#T5){ref-type="table"}). Our results support findings of previous studies by other workers, concerning the alterations in muscles and motor units after axotomy of peripheral nerves in neonatal rats. In the present study we concentrate on the time course of these alterations. We also focus on the influence of the in vivo administration of magnesium sulphate on motor unit survival and consequently on enhancing force outcome and muscle weight improvement. Administration of an NMDA or an AMPA receptor antagonist within this critical initial period of development, is thought to reverse the neurotoxic effects of axotomy and results in increased survival of motoneurons. Dizoscilpine malate (MK-801), an NMDA antagonist, has been used in animal models in vivowith success, in order to prevent motoneuron death after axotomy \[[@B15],[@B16]\]. However, it was badly tolerated by rats, due to side effects (high mortality). Magnesium is a non-competitive, voltage dependent, NMDA-receptor antagonist, acting by coupling with the specific Mg^2+^site within the pore of the ion channel \[[@B17]-[@B19]\]. Its similarity of action compared to MK-801 has been shown in two experimental models of neuropathic pain \[[@B20]\]. Moreover, axotomy in early postnatal period does not only reduce the number of surviving motoneurons and motor units, but also provokes changes in the contractile properties of limb muscles \[[@B21]\]. Immature muscle fibers have not gained yet the characteristics of fast- or slow- contracting type, since not all subtypes of the contractile proteins and enzymes have yet been formed. Extensor Digitorum Longus (EDL), for example, is normally a fast -- contracting and fatigable muscle in the adult rat. If axotomy is performed in neonates, EDL becomes slow and fatigue resistant. As other investigators have shown before \[[@B22]\], changes in the contractile properties of immature muscles, during normal development, go on for 30 days after birth, although establishment of mononeuronal innervation of muscle fibres is already fulfilled at P15 and the number of motoneurons innervating a specific muscle is constant after P0. It has been shown that the number of motor units remains constant after birth \[[@B14],[@B23]\]. This is consistent with our results: The EDL consisted of 40 and the soleus of 30 motor units in all age groups. It should be mentioned that soleus contractile properties are not significantly altered throughout early postnatal life. Soleus is a slow, non-fatigue resistant muscle at P14 that progressively becomes fatigue resistant. The process of soleus\' development into a fatigue resistance muscle is stopped, after sciatic nerve axotomy. Axotomized soleus becomes less fatigue resistant in adult rats, compared to control muscles. However the process of muscle necrosis, as proposed by other authors, could contribute to this result, as well, rather than the loss of motor units alone \[[@B25]\]. Denervated soleus muscle at birth is less fatigue resistant than control muscles in adult rats. This state is already established at P28, with this process progressing further on to adulthood. As mentioned already, it is noticeable that force outcome even by this \'conservative\' muscle is largely affected by axotomy as the animal grows up, after the first month of life. Our data show that even when no marked reduction in the number of motor units occurs, muscle weight does not improve from P28 to adulthood neither does single twitch or maximal tetanic tension. The same phenomenon appears in the EDL as well. The discrepancy seen during the P28 to adulthood interval, both in EDL and soleus, between the reduction in the number of motor units on the one hand, and force outcome (single twitch and tetanic twitch tension) and muscle weight after axotomy on the other, can be explained as a consequence of marked muscle atrophy and necrosis. As shown by other workers \[[@B24]\], all immature muscle fibers denervated at birth fail to become reinnervate, and the few reinnervated muscle fibers may be overloaded, hypertrophied and eventually necrotized. The final level of tension achieved by the denervated muscle, represents the equilibrium, between decrease in force due to atrophy and necrosis due to regeneration \[[@B25]\]. It has to be considered that nerve injury during early post-natal life, causes permanent changes in the muscles that are not caused by motoneuron death. Both immature EDL and soleus muscles are slow contracting at P14. During normal development EDL gradually converts into fast contracting, remaining not fatigue resistant throughout life. We found that EDL has already gained characteristics of a fast muscle at P21. Axotomy causing death of the bigger motoneurons, thus destroying the large motor units, converts the muscle into slow-contracting. It was also shown that increase in speed of both EDL and soleus was much reduced after denervation \[[@B26]\]. Our data show that EDL was affected by axotomy much more than soleus. If connection between muscle and nerve is disrupted, as is the case after axotomy, fastcontracting muscles are mostly impaired \[[@B12],[@B21],[@B27]\]. It is known that the overall poorer recovery of immature fast muscles after denervation seems to be due to preferential loss of fibers from fast motor units during reinnervation \[[@B28],[@B21],[@B29]\]. Histological findings \[[@B27]\] and studies on the isometric tension recordings by other workers \[[@B21]\], confirm our findings. However, there is one study \[[@B30]\], suggesting that soleus is the muscle predominately affected after nerve injury. Previous studies \[[@B6]\] have shown that sciatic nerve crush performed at 5--6 days after birth resulted in a 50% reduction in maximal tetanic tension of the EDL two months after injury, with the muscle becoming more fatigue resistant. Respective values for soleus, however, remained almost equal to those of control muscles. When nerve crush was performed at 3--5 days of life \[[@B7]\], single twitch tension values of the operated side were 36% of those observed for the control side. Respective value for maximal tetanic tension is 50% and for muscle weight is 61%. It is noticeable when comparing those data with our present results, that motoneurons at P2 are much more vulnerable to the excitotoxic effects of axotomy. Axotomy has also been shown to convert EDL into slow-contracting \[[@B7],[@B27]\]. On the other hand repeated injury to the sciatic nerve (at 5 and again at 11 days), has been shown to cause motoneuron death, mostly to the soleus, compared to the tibialis anterior & EDL motoneuron pool, in the spinal cord ventral horn \[[@B9]\]. Magnesium administration by other workers \[[@B31]\] has been shown to cause only a slight improvement in motoneuron survival after nerve crush at birth. However, the same authors found that daily in vivo administration of magnesium sulphate accompanied by NMDA, in axotomized rats at P5, rescues motoneurons destined to die. Our results are strongly suggestive that daily systemic treatment with magnesium sulphate, in order to keep sufficient blood concentrations of magnesium ions, results in increased motor unit survival. Conclusions =========== Our findings strongly support the findings of previous work \[[@B31]\], that has shown magnesium in vivo administration to rescue sciatic motoneurons from cell death, after axotomy. Local application of magnesium ions to the muscle, by implants affecting achetylcholine release in the neuromuscular junction \[[@B32]\], is suggested to reduce the number of surviving motoneurons by some other mechanism than blocking glutamate receptors. In our study however magnesium ions were not applied directly on the neuromuscular junction. It could be assumed that our results represent the resultant of magnesium ions actions on the nerve and the muscle. Magnesium administration did not cause any statistically significant influence on the contractile properties of the control muscles in the right unoperated leg, in any age group. Neuroprotection by magnesium reversed the effects of excitotoxicity, predominately in the fast-contracting EDL. In conclusion, our results show that motoneuron death occurs mostly within two weeks of axotomy, while systemic Mg^2+^administration rescues motoneurons and increases the number of motor units surviving into adulthood. Furthermore, fast and slow muscles respond differently to axotomy, as well as, to subsequent in vivo treatment with Mg^2+^. Competing interests =================== The authors declare that they have no competing interests. List of abbreviations ===================== AMPA: a-amino-3-hydro-5-methyl-4-isoxazolo-propionic acid CNS: Central Nervous System EDL: Extensor Digitorum Longus muscle FI: Fatigue Index HRT: Half Relaxation Time Mg: Magnesium MU: Motor Units NMDA: N-methyl-D-aspartate P: Postnatal SD: Standard Deviation TTP: Time to Peak Authors\' contributions ======================= NG carried out the experiments, participated in the sequence alignment as well as in the design of the study and drafted the manuscript. AH participated in the experiments and performed the statistical analysis DK participated in the experiments MA conceived of the study, and participated in its design and coordination Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2474/5/33/prepub> Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Time course of changes in the number of motor units of EDL and soleus muscles in the left hindlimb. The number of motor units in the right (control) hindlimb is set 40 in the EDL and 30 in the soleus. ::: ![](1471-2474-5-33-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Representative Isometric Tension and Fatigue Recordings of EDL and Soleus in adult rats ::: ![](1471-2474-5-33-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Number of Motor Units ::: Adults P28 P21 P14 ------------ --------------------- ------------ ----------- ----------- ----------- EDL Controls 40.00 37.83 39.50 39.33 ± 1.67 ± 1.33 ± 1.22 ± 1.21 Axotomy 7.50 10.83 11.50 13.17 ± 1.05 ± 1.94 ± 2.74 ± 2.14 Controls/Magnesium 39.83 41.00 38.67 37.50 ± 1.94 ± 0.63 ± 1.97 ± 2.74 Axotomy/ Magnesium 15.50 16.33 16.83 18.17 ± 1.76 ± 3.01 ± 2.71 ± 2.32 Soleus Controls 29.50 30.83 29.00 31.33 ± 1.64 ± 1.33 ± 1.79 ± 2.07 Axotomy 14.67 19.33 20.67 28.00 ± 1.75 ± 2.66 ± 1.63 ± 0.89 Controls/ Magnesium 30.00 29.50 28.83 29.83 ± 1.41 ± 2.35 ± 0.98 ± 2.40 Axotomy/ Magnesium 24.17 25.50 26.83 27.00 ± 1.72 ± 1.05 ± 1.72 ± 1.67 Values are expressed as Mean ± SD. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Muscle Weight, Number of Motor Units, Single Twitch and Maximal Tetanic Tension, expressed as a % ratio, of the operated to the control side, in the EDL, in the four age groups tested. Axotomy has been performed in P2, in all cases ::: EDL Operated / Control side % ------------------------------- --------- --------- -------- --------- Adult 4.63 3.31 18.80 10.60 ± 0.78 ± 0.30 ± 2.93 ± 2.62 Adult-Mg 16.59 15.16 38.84 38.88 ± 2.55 ± 0.89 ± 3.01 ± 5.25 P28 28.59 18.94 34.78 45.19 ± 4.58 ± 4.16 ± 3.03 ± 6.29 P28-Mg 38.77 40.14 39.84 64.07 ± 28.78 ± 19.34 ± 7.30 ± 7.22 P21 19.17 23.35 29.22 48.80 ± 10.78 ± 16.03 ± 7.44 ± 6.44 P21-Mg 58.64 64.22 43.73 58.08 ± 36.46 ± 43.02 ± 8.29 ± 29.92 P14 104.99 101.70 33.52 60.91 ± 55.27 ± 36.13 ± 5.60 ± 16.43 P14-Mg 65.94 83.51 48.51 66.11 ± 20.78 ± 33.10 18.80 ± 13.19 Values are expressed as Mean ± SD ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Muscle Weight, Number of Motor Units, Single Twitch and Maximal Tetanic Tension, expressed as a % ratio, of the operated to the control side, in the soleus, in the four age groups tested. Axotomy has been performed in P2, in all cases. ::: SoleusOperated / Control side % ----------------------------------- --------------- --------- ------------- --------------- Single Twitch Tet-100 Motor Units Muscle Weight Adult 16.80 12.44 49.87 10.60 ± 3.03 ± 0.97 ± 6.73 ± 2.62 Adult-Mg 87.34 97.00 80.56 38.88 ± 21.06 ± 11.33 ± 4.34 ± 5.25 P28 107.21 67.39 62.77 45.19 ± 53.62 ± 39.21 ± 8.87 ± 6.29 P28-Mg 94.10 91.45 86.67 64.07 ± 25.80 ± 15.09 ± 5.65 ± 7.22 P21 101.74 79.00 71.37 48.80 ± 17.13 ± 14.34 ± 5.45 ± 6.44 P21-Mg 93.49 77.01 93.74 58.08 ± 29.14 ± 25.63 ± 6.61 ± 29.92 P14 90.38 81.96 89.67 60.91 ± 22.35 ± 13.56 ± 6.41 ± 16.43 P14-Mg 89.40 92.12 49.87 66.11 ± 8.04 ± 11.15 ± 6.73 ± 13.19 Values are expressed as Mean ± SD. ::: ::: {#T4 .table-wrap} Table 4 AGE Condition Muscle Weight Single Twitch TET-10 TET-20 TET-40 TET-80 TET-100 ------------ --------------- ---------- ------------------- ------------------- ------------ ------------ ------------ ------------ ------------- ADULTS CON MEAN 0.163 67.74 92.83 117.50 172.33 212.17 259.50 SD 0.04 11.52 1.57 18.12 1.53 17.46 22.04 AX MEAN 0.017 3.08 2.87 3.16 6.39 7.67 8.54 SD 0.003 0.28 0.53 0.49 0.72 0.69 0.60 CON/MG MEAN 0.151 78.57 89.84 101.09 171.40 214.87 245.69 SD 0.008 9.80 10.60 11.12 1.39 17.92 22.69 AX/MG MEAN 0.059 12.84 13.06 15.79 19.40 27.76 37.31 SD 0.007 0.71 0.97 1.41 1.34 2.39 4.55 P28 CON MEAN 0.036 16.79 17.79 25.27 37.43 49.27 50.30 SD 0.004 2.74 2.38 2.87 5.51 5.22 5.23 AX MEAN 0.016 3.12 3.56 5.46 8.13 9.42 9.66 SD 0.003 1.10 1.19 2.34 2.72 2.80 2.81 CON/MG MEAN 0.016 13.39 14.08 21.90 28.24 31.29 31.61 SD 0.001 4.94 5.32 8.73 10.16 9.49 9.20 AX/MG MEAN 0.010 4.07 4.29 7.06 8.55 11.36 11.39 SD 0.001 1.11 0.95 2.11 1.95 1.09 1.07 P21 CON MEAN 0.021 9.09 9.64 13.98 19.00 21.75 21.65 SD 0.006 2.01 2.11 2.85 6.64 7.87 8.57 AX MEAN 0.010 1.80 1.98 2.87 4.42 4.88 4.98 SD 0.001 1.23 1.41 2.28 3.17 3.86 3.88 CON/MG MEAN 0.016 6.87 8.45 13.76 17.30 19.89 20.13 SD 0.005 2.97 4.85 7.54 9.76 12.12 13.24 AX/MG MEAN 0.008 3.39 3.72 6.47 8.42 9.42 9.47 SD 0.002 1.24 1.34 2.47 2.43 2.85 2.90 P14 CON MEAN 0.009 4.02 5.89 7.37 9.31 10.53 10.95 SD 0003 1.90 2.46 2.56 2.88 3.55 4.00 AX MEAN 0.006 3.48 4.94 7.12 8.86 10.07 10.04 SD 0.003 0.52 1.17 1.58 1.22 1.44 1.25 CON/MG MEAN 0.007 4.41 4.67 6.54 8.79 9.63 9.94 SD 0.002 0.83 0.87 0.73 2.10 2.59 2.59 AX/MG MEAN 0.005 2.85 4.76 5.92 8.15 7.63 7.65 SD 0.001 0.85 1.72 1.74 1.31 2.55 2.52 All values are presented in grams. TET-10, -20, -40, -80, -100, means Tetanic Twitch Tension in 10, 20, 40, 80, 100 Hz, respectively. ::: ::: {#T5 .table-wrap} Table 5 AGE Condition Muscle Weight Single Twitch TET-10 TET-20 TET-40 TET-80 TET-100 ------------ --------------- ---------- ------------------- ------------------- ------------ ------------ ------------ ------------ ------------- ADULTS CON MEAN 0.123 39.23 42.15 60.32 90,45 131.33 141.44 SD 0.015 7.84 5.22 15.30 8,31 16.95 13.37 AX MEAN 0.018 6.43 7.38 8.99 15,71 17.07 17.51 SD 0.003 0.65 0.52 0.89 1,61 1.01 0.90 CON/MG MEAN 0.106 37.05 51.00 71.75 99,32 130.48 138.02 SD 0.012 6.67 8.28 8.00 9,95 10.00 7.83 AX/MG MEAN 0.095 31.46 43.31 73.91 105,28 124.88 133.31 SD 0.010 4.41 6.27 9.73 9,98 11.62 10.60 P28 CON MEAN 0.029 8.45 10.37 16.45 23,42 27.34 29.07 SD 0.005 0.86 1.90 4.12 2,91 4.20 3.54 AX MEAN 0.019 9.24 10.06 11.94 16,16 18.25 19.61 SD 0.002 4.84 5.07 5.45 9,31 10.74 11.01 CON/MG MEAN 0.018 6.04 9.35 13.39 18,93 20.56 21.25 SD 0.004 1.91 2.69 2.56 1,28 3.19 3.01 AX/MG MEAN 0.014 5.37 7.09 7.81 14,11 18.73 19.35 SD 0.002 0.77 0.95 0.51 1,97 3.86 3.57 P21 CON MEAN 0.017 10.69 11.83 16.75 27,08 31.10 30.84 SD 0.002 1.42 1.90 2.24 3,30 4.67 4.54 AX MEAN 0.016 10.73 11.23 15.22 21,99 23.85 23.88 SD 0.001 1.22 2.49 1.90 2,91 2.60 1.75 CON/MG MEAN 0.014 5.89 7.69 10.83 19,70 23.64 23.24 SD 0.002 1.69 2.39 3.22 5,90 9.27 8.69 AX/MG MEAN 0.010 5.26 6.42 12.22 14,89 16.94 17.54 SD 0.003 1.25 2.32 3.51 4,00 4.56 4.65 P14 CON MEAN 0.006 4.01 6.16 7.32 9,14 10.33 10.46 SD 0.001 1.62 1.20 1.48 1,96 2.10 2.02 AX MEAN 0.005 3.35 5.72 6.79 7,59 8.18 8.35 SD 0.001 0.61 0.88 0.57 0,32 0.57 0.60 CON/MG MEAN 0.006 3.61 4.16 6.37 9,16 11.15 11.35 SD 0.002 0.86 1.16 3.10 5,05 4.85 4.62 AX/MG MEAN 0.005 3.18 3.53 5.98 7,97 9.79 10.06 SD 0.000 0.52 0.54 1.12 1,54 2.71 2.94 All values are presented in grams. TET-10, -20, -40, -80, -100, means Tetanic Twitch Tension in 10, 20, 40, 80, 100 Hz, respectively. ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Contraction (TTP) -- Relaxation (HRT) Time (msec) ::: EDL ------------ ------------ ----------- ------------ ------------ ------------ ------------ ----------- ------------ Control 56 ± 4.97 59 ± 4.34 38 ± 6.45 42 ± 14.61 35 ± 3.93 32 ± 2.34 36 ± 4.40 28 ± 2.28 Axotomy 56 ± 4.97 60 ± 6.25 49 ± 10.18 56 ± 16.26 43 ± 5.48 44 ± 7.92 77 ± 7.89 71 ± 11.50 Control/Mg 54 ± 4.97 56 ± 4.08 40 ± 2.83 34 ± 7.20 44 ± 2.81 38 ± 4.80 40 ± 4.90 34 ± 3.20 Axotomy/Mg 56 ± 4.49 62 ± 4.80 49 ± 3.72 46 ± 10.61 43 ± 3.72 41 ± 7.34 38 ± 7.53 43 ± 4.13 SOLEUS P14 P21 P28 ADULT TTP HRT TTP HRT TTP HRT TTP HRT Control 74 ± 8.29 66 ± 8.00 58 ± 5.22 64 ± 5.85 48 ± 3.25 58 ± 3.70 56 ± 3.37 60 ± 3.78 Axotomy 78 ± 6.81 75 ± 5.75 60 ± 9.77 63 ± 7.39 51 ± 7.45 58 ± 10.90 58 ± 5.99 61 ± 7.23 Control/Mg 75 ± 5.90 76 ± 6.62 51 ± 14.72 53 ± 27.62 49 ± 13.20 61 ± 13.41 56 ± 5.47 60 ± 7.55 Axotomy/Mg 73 ± 6.54 72 ± 6.12 54 ± 2.53 47 ± 4.34 59 ± 5.22 53 ± 8.14 61 ± 2.00 60 ± 3.77 Values are expressed as Mean ± SD. ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Fatigue Index ::: Fatigue Index % -------- --------- ----------------- ------ ------ ------ EDL Control 65.4 61.4 75.1 55.0 Axotomy 57.2 44.7 41.6 15.6 Con/Mg 72.4 72.4 63.1 56.4 Axot/Mg 65.2 40.6 34.2 9.9 Soleus Control 55.6 42.0 35.8 17.8 Axotomy 68.3 42.7 44.8 34.7 Con/Mg 67.1 34.5 30.6 19.1 Axot/Mg 64.9 54.0 58.9 19.8 Values are expressed as Mean ± SD. :::	PubMed Central	2024-06-05T03:55:48.443996	2004-9-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522819/", "journal": "BMC Musculoskelet Disord. 2004 Sep 24; 5:33", "authors": [ { "first": "N", "last": "Gougoulias" }, { "first": "A", "last": "Hatzisotiriou" }, { "first": "D", "last": "Kapoukranidou" }, { "first": "M", "last": "Albani" } ] }
PMC522820	Background ========== Radiotherapy is widely used in treating different types of cancer and is an effective therapeutic modality against abdominal and pelvic cancers. Gastrointestinal tract damage by radiotherapy limits its efficacy in cancer treatment. The small bowel is highly radiosensitive and very mobile and is thus an important dose-limiting organ during radiation therapy for abdominal and pelvic cancer \[[@B1]\]. Radiation induces an inflammatory response in target and surrounding tissues, which is characterised by accumulation of plasma proteins and leucocytes. Leucocyte recruitment is a multi-step process, which includes leucocyte rolling, activation and firm adhesion to the endothelium \[[@B2]\]. Leucocyte rolling reduces the velocity and allows time for leucocytes to detect chemotactic signals on the endothelial surface. It is now widely accepted that leucocyte rolling is a precondition for the subsequent firm adhesion and extravascular accumulation \[[@B3],[@B4]\]. With intravital microscopy, several studies have demonstrated that the selectin family of adhesion molecules predominantly mediates leucocyte rolling and that stationary adhesion is supported by the β~2~-integrins \[[@B5],[@B6]\]. Several animal models exist in order to investigate radiation-induced leucocyte endothelium cell responses, which can broadly be divided into two groups. Topical radiation (abdominal and pelvic) \[[@B7]-[@B9]\] and more recently, segmental radiation of an isolated short segment of small intestine where different responses to radiation are examined \[[@B1],[@B10]\]. We refined and developed the latter model, incorporating a platform where mice can be placed allowing exposure of the required segment of intestine for irradiation. This allows us to expose exteriorised intestinal sections to tailored high dose radiation, greatly minimizing scattering effects and thereby consequently avoiding surrounding tissue damage. The purpose of this study was to refine a small bowel radiation model in order to study the time responses in terms of leucocyte rolling, adhesion, myeloperoxidase (MPO) levels, histology and intestinal floral changes in response to high dose radiation of the ileum, where the exact biologically effective dose could be calculated. Methods ======= Animals ------- Male C57Bl/6J mice weighing 22--26 g were kept under standard laboratory conditions maintained on a 12 hour light and 12 hour dark cycle and were allowed free access to animal chow and tap water ad libitum. All experimental procedures were performed in accordance with legislation on the protection of animals and were reviewed and approved by the Lund University Ethic\'s Committee for Animal Experimentation. Anesthetic and surgical preparation ----------------------------------- The mice were anesthesized with 7.5 mg Ketamine hydrochloride (Hoffman-La Roche, Basel Switzerland) and 2.5 mg Xylazine (Janssen Pharmaceutica, Beerse, Belgium) per 100 g body weight by intraperitoneal (i.p.) injection. The animals were placed in supine position on a heating pad (37°C) for maintenance of body temperature. A small midline incision (1.0--1.5 cm) was performed and a 5 cm segment of ileum located 5 cm from the ileocaecal valve was exteriorised and marked with 5-0 non-absorbable sutures. Any other visible prolapsed abdominal content was replaced back into the abdomen and the animal was placed on the specially designed frame/chamber (Figure [1](#F1){ref-type="fig"}), with the loop of intestine fixed between two perspex sheets. The exposed ileum was subjected to a single dose of high dose radiation of 19 Gy and thereafter replaced in the abdomen and the incision closed with a polypropylene suture. At the appropriate time a polyethylene catheter (PE-10 with an internal diameter of 0.28 mm) was placed into the internal jugular vein for administration of fluorescent markers. Leucocyte-endothelium interactions were then observed using an inverted intravital fluorescence microscopy (IIVM) at different time points (2--48 hours). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The chamber for segmental intestinal radiation exposure. ::: ![](1471-2482-4-10-1) ::: Experimental protocol --------------------- The animals were divided into two different groups; Radiation & Surgery group (R+) and Sham Radiation & Surgery group (R-) which served as negative controls. The irradiated groups of mice (n= 6/time point) were exposed to 19 Gy of radiation and leucocyte-endothelium interactions were measured 2, 6, 16, 24 and 48 hours after induction of radiation using an IIVM (each group n= 6, likewise the R- groups). At the end of the procedure samples were collected for histology, and measurement of intestinal microflora, MPO and systemic leukocyte counts. Radiation --------- The irradiations were undertaken using a clinical linear accelerator (Varian Clinac 2100C). The exteriorised mouse intestine was positioned between Perspex slabs to accomplish sufficient secondary radiation scatter and thereby a reproducible and homogenous dose distribution. The absorbed dose was verified with independent measurements and was found to be within 5% throughout the intended volume using this technique. Using an asymmetrically half blocked 6 MV beam and extra lead shielding (Figure [1](#F1){ref-type="fig"}), the treatment field perfectly fitted the exteriorised intestine while the remaining body was kept outside the radiation beam. An absorbed dose of 19 Gy was delivered to the intestine as this dose causes consistent structural, cellular, and molecular changes \[[@B11]\]. The absorbed dose rate was 3.2 Gy/minute and consequently the irradiation time for each animal was approximately 6 minutes. During irradiation the intestine in the chamber is protected from large temperature variations and trauma by perspex sheets. The exposure time from surgery, through irradiation to wound closure is kept at a minimum, taking approximately 15 minutes, thus keeping stress and trauma levels low. Intravital microscopy --------------------- Observations of the intestinal microcirculation were made using an inverted Olympus microscope (IX70, Olympus Optical Co. GmbH, Hamburg Germany) equipped with different lenses (x10/NA 0.25 and x40/NA 0.60). The microscopic images were televised using a charge-coupled device video-camera (FK 6990 Cohu, Pieper GmbH, Schwerte, Germany) and recorded on videotape (Sony SVT-S3000P S-VHS recorder) for subsequent off-line analysis. To prevent drying during microscopic observations the intestinal segment was placed on a saline moistened cotton gauze and thereafter positioned under the microscope. After a 5-min equilibration period, quantitative measurements were taken. Analysis of leucocyte-endothelium interactions (rolling and adhesion) was made in venules (inner diameter 15--30 μm) with stable resting blood flow. Blood perfusion within individual microvessels was studied after contrast enhancement by i.v. administration of fluorescein isothiocyanate (FITC)-labelled dextran (MW 150000), (0.05 ml, 5 mg/ml, Sigma Chemical Co. St. Louis, MO, U.S.A.). In vivolabelling of leucocytes with rhodamine-6G (0.1 ml, 0.5 mg/ml, Sigma Chemical Co. St. Louis, MO, U.S.A.) enabled quantitative analysis of leucocyte flow behavior in the ileum microcirculation. Due to its relatively higher molecular weight FITC-dextran stains/labels the intravascular plasmatic phase of the blood under epi-illumination with blue light (excitation wavelength 490 nm; emission wavelength 510 nm) whereas the lower molecular weight of Rhodamine 6G allows for labelling of leucocytes and platelets using green fluorescent light (excitation wavelength 530 nm; emission wavelength 560 nm). Quantification of microcirculatory parameters was performed off-line by frame-to-frame analysis of the videotaped images. Leucocyte rolling was determined by counting the number of leucocytes passing a reference point in the venule per 20 sec and is expressed as cells/min. Firm adhesion was measured by counting the number of cells adhering to the venular endothelium (200--300 μm long segments) and remained stationary for 20 sec and is given as cells/mm venule length. Blood flow velocities were analysed by means of a video assisted computer image analysis programme, CapImage software (Zeintl, Heidelberg, Germany). The staining of the plasmatic phase by FITC-dextran gives an indirect enhancement of red blood cells which appear dark in the illuminated surrounding plasma. The CapImage uses the FITC-dextran image to calculate the red blood cell velocity. The velocity was calculated as a mean value from 5--8 measurements per venule and is expressed as mm/sec. Venular wall shear rate was determined based on the Newtonian definition: wall shear rate = 8 \[(red blood cell velocity/1.6)/venular diameter\] as described previously \[[@B12]\]. MPO measurement --------------- The enzyme myeloperoxidase (MPO) is abundant in neutrophil leucocytes and has been found to be a reliable marker for the detection of neutrophil accumulation in inflamed tissue. To determine tissue MPO content, radiated ileal tissue was collected, weighed, homogenized in 10 ml 0.5% hexadecyltrimethylammonium bromide, and freeze thawed, after which the MPO activity of the supernatant was assessed. The enzyme activity was determined spectrophotometrically as the MPO-catalysed change in absorbance occurring in the redox reaction of H~2~O~2~(460 nm, 25°C). Values are expressed as MPO units per g tissue. Histological study ------------------ Samples from the irradiated small intestine were placed in 4% phosphate buffered formaldehyde. Paraffin-embedded samples were sliced and studied under light microscopy after staining with hematoxylin and eosin. At least 3 slides were studied from each specimen in a blinded fashion. Intestinal microflora --------------------- Tissue samples from the irradiated small intestine were first placed in 5 ml of sterile transport medium \[[@B13]\]. Samples were then placed in an ultrasonic bath (Millipore, Sweden) for 5 minutes and then rotated on Chiltern (Terma-Glas, Gothenberg, Sweden) for 2 minutes. After a conventional dilution procedure, viable counts were obtained from Brain Heart Infusion (BHI) that was incubated aerobically and anaerobically at 37°C for 72 hours (aerobic and anaerobic bacterial count, respectively), and from Rogosa agar (Oxoid, Hampshire, England) that was incubated anaerobically at 37°C for 72 hours (lactobacilli counts). Viable counts were also obtained from violet red-bile-glucose agar (VRBD) (Oxoid, Hampshire, England) that was incubated aerobically at 37°C for 24 hours (Enterobacteriaceae counts) and from BHI agar containing gram-negative anaerobic supplement (Oxoid, Hampshire, England) that was incubated anaerobically at 37°C for 72 hours (gram negative anaerobic bacterial counts). Systemic leucocyte counts ------------------------- 20 μl blood was mixed with Turk\'s solution (0.2 mg gentian violet in 1 ml glacial acetic acid, 6.25 % v/v) in a 1:10 dilution. Leucocytes were counted and differentiated as polymorphonuclear (PMNL) or mononuclear (MNL) cells in a Burker chamber. Statistical analysis -------------------- Statistical evaluations were performed using the Kruskal-Wallis one way analysis of variance on ranks for unpaired samples (Dunn\'s post hoc test was used). For bacterial microflora in comparing 2 groups we used Mann-Whitney Rank sum test, and for the comparison of the different time points within the radiated groups we used One Way ANOVA followed by multiple comparisons versus control group (Dunnett\'s method). The results are presented as mean values ± SEM. Differences were considered to be significant at P\< 0.05. Results ======= Radiation-induced leucocyte-endothelium interactions in the ileum ----------------------------------------------------------------- Intravital microscopic studies in post-capillary venules of the distal ileum in sham operated mice (controls) revealed only occasional interactions between leucocytes and the microvascular endothelium, i.e. the number of rolling and adherent leucocytes was 2.4 ± 1.2 cells/min and 1.7 ± 1.7 cells/mm, respectively. In contrast, radiation (19 Gy) evoked a marked time-dependent leucocyte response, i.e. a significant increase in both leucocyte rolling and firm adhesion over time (Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"}, P\< 0.05, vs. controls, n= 5--10). We observed that leucocyte rolling peaked two hours after radiation (38 ± 7 cells/min, (Figure [2](#F2){ref-type="fig"}), P\< 0.05 vs. sham, n= 6--10), whereas leucocyte adhesion was maximum after 16 hours showing a marked response of 59 ± 14 cells/mm (Figure [3](#F3){ref-type="fig"}, P\< 0.05 vs. sham, n= 6). Interestingly, both the leucocyte rolling and adhesion responses to radiation returned to baseline levels 48 hours after radiation (Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"}, P\> 0.05, vs. sham, n= 5--10). There was no difference in the hemodynamic parameters between the different experimental groups (Table [1](#T1){ref-type="table"}) and also no significance difference could be seen in the systemic leucocyte counts. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Venular leucocyte rolling in the mouse ileum at different time points after radiation. Data represents mean ± SEM. ::: ![](1471-2482-4-10-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Venular leucocyte adhesion in the mouse ileum at different time points after radiation. Data represents mean ± SEM. ::: ![](1471-2482-4-10-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Hemodynamic parameters in ileal venules ::: *Diameter(μm) Red blood cell velocity (mm s^-1^) Wall shear rate (s^-1^) ------------------ -------------------- ------------------------------------ ------------------------- Sham 2 hrs 25.5 ± 2.1 1.57 ± 0.21 314 ± 44 Sham 16 hrs 26.2 ± 1.2 1.61 ± 0.14 309 ± 19 Radiation 2 hrs 22.2 ± 1.6 0.93 ± 0.08 212 ± 44 Radiation 6 hrs 24.5 ± 2.4 1.53 ± 0.12 319 ± 56 Radiation 16 hrs 21.7 ± 2.5 0.93 ± 0.22 222 ± 74 Radiation 24 hrs 20 ± 1.9 1.26 ± 0.16 408 ± 48 Radiation 48 hrs 25.5 ± 2.3 1.32 ± 0.11 264 ± 57 Radiation was directed to the ileum and leukocytes responses were measured after 2--48 hours. Sham-operated controls underwent identical procedures except undergoing radiation. Responses measured at 2 and 16 hours (n = 5,6). Blood flow velocities were measured off-line by frame-to-frame analysis of the videotaped images. Data are mean ± SEM. ::: Histological changes following radiotherapy ------------------------------------------- At 2 hours we could not observe any marked differences in the number of inflammatory cell types compared to the controls (Figure [4](#F4){ref-type="fig"}). At 6 hours we found quite a number of apoptotic epithelial cells, and a few inflammatory cells -- mainly neutrophil granulocytes in the lamina propria. Both the granulocytes and the apoptotic cells increased in numbers at 16 hours. An increase in the inflammatory infiltrate was also observed in the smooth muscle layer (muscularis propria). 24 hours after radiation the muscularis mucosae was oedematous and infiltrated by granulocytes; there was clearly visible lymph vessel ectasia and apoptosis mainly in the deeper parts of the crypts. Forty-eight hours after radiation a vast increase of goblet and apoptotic cells was seen in the whole length of the epithelium and crypts. On the other hand there was a reduction in lymph vessel ectasia, oedema and in the number of inflammatory cells present (Figure [5](#F5){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### ACross section of intestinal wall 2 hrs after irradiation. No marked differences in the number of inflammatory cell types compared to the controls. BCross section of intestinal wall 48 hrs after irradiation. A vast increase of goblet and apoptotic cells was seen in the whole length of the epithelium and crypts. There was a reduction in lymph vessel ectasia, oedema and in the number of inflammatory cells present compared to earlier time points. ::: ![](1471-2482-4-10-4) ::: Intestinal microflora --------------------- Compared to the sham groups; the aerobic, Enterobacteriaceae, Lactobacillusand anaerobic counts had decreased two hours after radiation (Figure [5](#F5){ref-type="fig"}), the same groups, with the exception of the anaerobic count were significantly decreased sixteen hours after radiation (Figure [6](#F6){ref-type="fig"}). There were no significant differences between the experimental groups twenty four hours after radiation compared to the sham group (Figure [7](#F7){ref-type="fig"}). When assessing the trends within the various radiated bacterial groups compared to the 24 hour levels we found significant decreases in the aerobic count at 2 hours; in the anaerobic count at 2 and 6 hours; and in the Enterobacteriaceaeat 2, 6 and 16 hours. There were no significant changes in the Lactobacilluscount at the different time points within the radiated groups (Figure [8](#F8){ref-type="fig"}). ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Ileum bacterial microflora in sham and 2 hours after radiation groups. \ denotes p \< 0.05 compared to sham group. ::: ![](1471-2482-4-10-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Ileum bacterial microflora in sham and 16 hours after radiation groups. \* denotes p \< 0.05 compared to sham group. ::: ![](1471-2482-4-10-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Ileum bacterial microflora in sham and 24 hours after radiation groups. No significant difference between the experimental groups. ::: ![](1471-2482-4-10-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Ileum bacterial microflora in the radiated groups at different time points. \* denotes p \< 0.05 compared to 24 hours radiated group. ::: ![](1471-2482-4-10-8) ::: MPO measurement --------------- There were no differences in MPO measurements in the experimental groups. Discussion ========== The frequent use of radiotherapy for abdominal and pelvic malignancies results in an increased risk of radiation enteritis \[[@B14]\]. The dose of radiation that can be applied in clinical practice is usually limited by the need to restrict the number and severity of side effects in normal tissues surrounding a tumour, which are unavoidably exposed to radiation \[[@B8]\]. Intestinal radiation toxicity (radiation enteropathy) is characterised by mucosal barrier breakdown and inflammation, followed by development of progressive vascular sclerosis and intestinal wall fibrosis. The process is accompanied by sustained over expression of inflammatory and fibrogenic cytokines \[[@B15],[@B16]\]. An early inflammatory response, beginning a few hours after irradiation, characterised by leucocyte infiltration into the irradiated organs is regarded as one of the main determinants of radiation-induced organ damage \[[@B17],[@B18]\]. The development of an inflammatory response involves sequential leucocyte-endothelial cell interactions. Different families of cell adhesion molecules have been shown to participate in the process of leucocyte recruitment \[[@B19]\]. There are three major families of adhesion molecules involved in the leucocyte recruitment process, the selectins, the integrins and the immunoglobulin supergene families \[[@B20]\]. The present study has concentrated on the acute effects of radiation injury on leucocyte rolling and adhesion at specific time points after radiation. We found that radiation evoked a marked time dependent leucocyte response with a significant increase in both leucocyte rolling and firm adhesion over time. Leucocyte rolling peaked 2 hours after radiation whereas leucocyte adhesion was highest after 16 hours showing a marked response. Interestingly, both the leucocyte rolling and adhesion responses to radiation were back to baseline 48 hours after radiation. An intravital microscopic study of radiation-induced leucocyte-endothelial cell interaction using abdominal radiation and a dose of 20 Gy revealed an increased leucocyte rolling in mesenteric venules 2 hours after radiation, with a marked increase in leucocyte adhesion and emigration noted at 6 hours \[[@B18]\]. In another study of radiation-induced inflammatory damage, abdominal irradiation was administered using 4 and 10 Gy respectively \[[@B8]\]. Here an increase in leucocyte rolling was observed 2 hours after radiation, which then returned to basal levels at 6 and 24 hours respectively. An increase in leucocyte adhesion was also observed 2 hours after irradiation, which was then sustained during the 24 hour observation period \[[@B8]\]. In our study we showed the maximum effect on rolling after 2 hours and adhesion after 16 hours and the return to basal levels 48 hours after radiation. We used a single high dose radiation of 19 Gy directly to an exteriorised segment of ileum. This dose was chosen because it has been shown to give a good correlation or dose response relationship of histopathological changes (e.g. mucosal ulceration, vascular sclerosis) to incidence of clinical complications and cellular evidence of injury \[[@B1]\]. When comparing our results (ileal venule measurements) to those from other tissue, namely from the pial venules of cerebral microvasculature of the rat after 20 Gy irradiation \[[@B21]\], we find that the results follow a similar time course. Assuming that the radiation dose distribution is similar in all experiments mentioned above, the differences in peak times for leucocyte rolling and adhesion may probably be due to differences in radiation dose/duration, the extent of trauma, the effect of anaesthesia, the mode and duration of experiments. Endogenous bacterial flora produces nutrients (e.g. short-chain fatty acids) for the mucosa; prevents overgrowth of potentially pathogenic micro-organisms; stimulates the immune system especially the gut-associated lymphoid tissue; helps eliminate toxins from the lumen and participates in intestinal regulation, motility and blood flow \[[@B22]\]. Radiation on the other hand influences and alters the mucosal microflora, and this in combination with barrier dysfunction leads to a translocation of microbes through the mucosa into blood circulation \[[@B23]\]. Our experiment shows that radiation affects the intestinal microflora. Two hours after radiation the aerobic, anaerobic, Enterobacteriaceaeand Lactobacilluscounts were decreased and after 16 hours the aerobic, Enterobacteriaceaeand Lactobacilluscounts were still decreased in the radiated groups compared to sham controls. Twenty-four hours after radiation there was no significant difference between the experimental groups. Comparing the results of the irradiated groups alone, one observes an increase in bacterial count over time after radiation. It seems that radiation decreases the bacterial count at early time points with no difference in total bacterial count at late time points. This total count does not reflect the difference in bacterial species within each group, and thus, further investigations are needed to study the imbalances that occur. One study has shown that microorganisms such as Escherichia, Proteus, Clostridium, normally absent in healthy animals, appear in the intestines of guinea pigs subjected to irradiation. At the same time lactobacilli and bifidobacteria sharply decrease in number \[[@B24]\]. Bacterial overgrowth and intestinal pseudo-obstruction may succeed abdominal radiotherapy and the impaired motility emerges as a causal factor for gastrointestinal colonization with gram-negative bacilli. Abnormal motility and gram-negative bacilli in the gut may be essential in the pathogenesis of late radiation enteropathy \[[@B25]\]. Changes in intestinal microflora therefore most probably affect the course of the development of radiation enteropathy. Acute intestinal symptoms during pelvic radiotherapy may not depend only on mucosal damage \[[@B26]\]. Post-radiation gut structural damage occurs early and parallels functional changes of the intestinal mucosa, including increased epithelial permeability (shown both in vivo and ex vivo), activation of secretory pathways, decreased nutrient absorption, diarrhoea, and weight loss \[[@B27]\]. The microfloral changes, which we have shown, could play an important role in the structural and functional intestinal changes after radiation, particularly in the presence of intestinal mucosal changes and increased intestinal permeability. Patients with carcinoma of the uterine cervix or endometrium receiving postoperative radiation therapy have a significant decrease in intestinal microflora after the first radiation exposure, whereas at the end of radiotherapy all bacteria have increased and reached basal values except Enterococcus faecium1, lactobacilli and total anaerobes. In some patients an overgrowth of some Clostridium spp. (potential pathogens) associated with clinical symptoms, was observed. Patients receiving radiotherapy may thus benefit from the intake of oral bacteriotherapy \[[@B28]\]. The importance of investigating the effects of radiation on the different bacterial species within the total count is therefore of significance for the modulation of treatment regimes. The histological changes following radiation are both time and dose dependent \[[@B29],[@B30]\]. Soon after radiotherapy we observed an increase in inflammatory cell-infiltrate, apoptosis, mucin producing goblet cells and oedema, representing the morphological expression of an unspecific reactive process with a supposed protective function. Variations of these changes have been previously observed in the clinical situation. The vast increase in goblet cells that we observed may resemble that seen in necrotising enterocolitis. A resemblance to chronic idiopathic inflammatory bowel disease, eosinophilic colitis and microscopic colitis can also be seen if the mild crypt distortion or withering that occurs with radiation injury is confused with proper crypt architectural distortion of inflammatory disease. Isolated crypts due to nuclear regenerative changes may also mimic the microadenomas of familial adenomatosis polyposis \[[@B30]\]. Histological changes in the pre-existing normal mucosa following preoperative radiotherapy need to be appreciated by the histopathologist if we are to avoid erroneous concurrent diagnosis \[[@B30]\]. Furthermore, a correct assessment of the effects of new treatment regimes or prophylaxis is based on a sound histological judgment. No differences MPO values could be seen between the controls and the radiated groups. This is probably because it is a crude method of measurement and thus may not be sensitive enough to detect early changes of inflammation. This study therefore presents a refinement of previous methods of examining effects of radiation enteropathy, and a new approach at investigating radiation induced leucocyte responses in the ileal microcirculation. This new model may be instrumental in developing strategies against pathological recruitment of leucocytes and changes in intestinal microflora in the small bowel. Competing interests =================== None declared. Authors\' contributions ======================= LBJ designed the study and participated in construction of the chamber. Performed experimental studies and drafted the manuscript. AAR performed experimental studies and drafted the manuscript. DA participated in the design of the study and construction of the chamber. Performed experimental studies, drafted the manuscript and performed the statistical analysis. LW participated in the radiological design of the study, construction of the chamber and the implementation of radiotherapy. SB participated in the radiological design of the study, chamber and the implementation of radiotherapy. CT participated in the implementation of radiotherapy. NO carried out bacteriological studies. VC performed the histological analysis. HT assisted with issues related to intravital microscopy. BJ conceived of the design, participation in construction of the chamber, co-ordination of the study as well as supervision and draft of the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2482/4/10/prepub> Acknowledgements ================ This study was supported by grants from Cancer Foundation of Sweden, Dir. A. Påhlsson\'s Foundation, Ruth & Richard Juhlin\'s Foundation, Malmö University Hospital, Lundgren\'s Foundation, Gunnar Nilsson\'s Foundation, Royal College of Surgeons International Fellowship, Apotekaren Hedberg\'s Fond, and Einar & Inga Nilsson\'s Foundation.	PubMed Central	2024-06-05T03:55:48.449140	2004-9-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522820/", "journal": "BMC Surg. 2004 Sep 13; 4:10", "authors": [ { "first": "Louis Banka", "last": "Johnson" }, { "first": "Amjid Ali", "last": "Riaz" }, { "first": "Diya", "last": "Adawi" }, { "first": "Lena", "last": "Wittgren" }, { "first": "Sven", "last": "Bäck" }, { "first": "Charlotte", "last": "Thornberg" }, { "first": "Nadia", "last": "Osman" }, { "first": "Virgil", "last": "Gadaleanu" }, { "first": "Henrik", "last": "Thorlacius" }, { "first": "Bengt", "last": "Jeppsson" } ] }
PMC522821	Background ========== Cytotoxic T-lymphocytes (CTL) and natural killer (NK) cells play an important role in host defense against intracellular pathogens and tumor cells. CTL recognize target cells through processed antigenic peptides presented via MHC. In contrast, NK cells mediate lysis of numerous cellular targets without classical MHC restriction. NK cells appear to use a variety of different, non-rearranging receptors to initiate cytoxicity and cytokine production \[[@B1]\]. Although CTL and NK differ in the receptors they use to recognize target cells, they both utilize the granule exocytosis and the Fas ligand (FasL)-mediated pathways to eliminate altered-self targets \[[@B2]-[@B6]\]. The granule-mediated pathway is the dominant pathway in CTL and NK \[[@B5]\]. CTL and NK cell granules contain a number of proteins, including perforin and granzymes, with GrB being the most abundant granzyme present \[[@B7],[@B8]\]. Upon recognition and conjunction of the effector cell with the target, preformed granules containing GrB polarize in cytolytic lymphocytes at the point of contact and are secreted into the intercellular space formed between the effector and target cell \[[@B9]-[@B14]\]. The secretion of GrB occurs quite rapidly, is Ca^2+^-dependent, and mediates the lethal hit that kills virus-infected and tumor cells \[[@B2],[@B7],[@B8],[@B10],[@B15]-[@B19]\]. Cell-mediated cytotoxicity has conventionally been measured using the standard ^51^Cr-release assay \[[@B20]\]. Recently, the use of the IFN-γ ELISPOT assay as a surrogate measure for CTL and NK responses has gained increased application. However, the IFN-γ ELISPOT assay may not be an accurate measure of cytotoxic lymphocytes as non-cytotoxic cells can secrete IFN-γ. Since GrB is exclusively present in the granules of CTL and NK cells and is a key mediator of the granule exocytosis-mediated cytolytic pathway \[[@B21]-[@B23]\], it is an excellent candidate marker for immunological monitoring of innate immunity by the ELISPOT method. The ELISPOT method has been successfully applied to measure GrB secretion by GrB-transfected CHO cells and for assessing antigen specific T-cell cytotoxic activity \[[@B24],[@B25]\]. In this study, we utilized human NK-like and lymphokine-activated killer (LAK) effector cells to assess whether the GrB ELISPOT assay could accurately measure the MHC non-restricted cytolysis that occurs upon recognition of appropriate target cell ligands by activating receptors on these effector cells. Additionally, we evaluated whether the ELISPOT assay measured GrB release due to degranulation of stimulated effector cells. The GrB ELISPOT assay was strategically compared to the IFN-γ ELISPOT and the standard ^51^Cr-release assays to determine if the GrB ELISPOT assay is a viable or better alternative to measure innate immunity. Methods ======= Target cell lines ----------------- K562 cells (Human myelogenous leukemia cell line, ATCC, Manassas, VA) were cultured at 37°C, 5% CO~2~in tissue culture medium (TCM) consisting of RPMI 1640 (BioWhittaker, Walkersville, MD) supplemented with 10% fetal bovine serum (FBS; Hyclone, Logan, UT), 1 mM non-essential amino acids, 2 mM glutamine, 100 U/ml Penicillin, 100 μg/ml Streptomycin, 20 mM HEPES and 1 mM sodium pyruvate (Gibco BRL Life Technologies, Grand Island, NY). TALL-104 effector cells ----------------------- The TALL-104 cell line is a T-cell derived line that is highly cytotoxic for NK-sensitive targets and when grown in the presence of IL-2, can be stimulated to secrete IFN-γ and GrB \[[@B26],[@B27]\]. The cell line was cultured in Iscove\'s modified Dulbecco\'s medium (BioWhittaker) supplemented with 20% FBS (Hyclone), 4 mM glutamine, 100 U/ml Penicillin and 100 μg/ml Streptomycin (Gibco BRL Life Technologies) at 8% to 10% CO~2~. Recombinant human IL-2 (100 U/ml; Hoffmann-LaRoche, Nutley, NJ) was added every 2--3 days to ensure optimal growth and maintenance of cytotoxic activity. Generation of human lymphokine-activated killer (LAK) cells ----------------------------------------------------------- Peripheral blood mononuclear cells (PBMC) were isolated from venous blood of normal human volunteers by buoyant density centrifugation over Ficoll-Paque (Pharmacia, Piscataway, NJ). Aliquots of effector cells were cryopreserved in the vapor phase of liquid N~2~for future use in functional testing. The PBMC were thawed and resuspended at 2 × 10^6^cells/ml in 20 ml of RPMI 1640 (BioWhittaker) containing 10% human AB serum (Mediatech, Herndon, VA), 1 mM non-essential amino acids, 2 mM glutamine, 1 mM pyruvate, 20 mM HEPES, 100 U/ml Penicillin and 100 μg/ml Streptomycin (Gibco BRL Life Technologies). Cell suspensions were stimulated with 100 U/ml of IL-2 (Hoffmann-LaRoche) on day 0 and cultured for 5 to 6 days at 37°C, 5% CO~2~. LAK cultures consisted of 30.0 ± 2.1 % NK cells (CD3^-^,CD16/CD56^+^) and 26.9 ± 4.9 CD8^+^T-cells as determined by flow cytometric analysis. Secretion Inhibitors -------------------- TALL-104 cells were treated with inhibitors of cellular secretion prior to use in the GrB and IFN-γ ELISPOT assays. Inhibitors used included 1,2-bis-(2-aminophenoxy)ethane-N,N,N\', N\'-tetraacetic acid tetra(acetoxymethyl) ester (BAPTA-AM, Molecular Probes, Eugene, OR), a cell permeant chelator of intracellular Ca^2+^, and γ,4-dihydroxy-2-\[6-hydroxy-1-hep-tenyl\]-4-cyclopentanecrotonic acid λ-lactone (brefeldin A; Sigma, St. Louis, MO) which blocks protein secretion. TALL-104 cells were resuspended at 2.5 × 10^5^cells/ml and 2.5 × 10^4^cells/ml in PBS without divalent cations for BAPTA-AM and brefeldin A treatment, respectively. Cells were pretreated with the indicated concentrations of BAPTA-AM for 45 min or with 5 μg/ml brefeldin A for 1 h at 37°C, washed twice, assessed for viability by trypan blue exclusion, and resuspended at 2.5 × 10^4^cells/ml in assay media. ^51^Cr-release assay -------------------- Cytotoxicity of TALL-104 and LAK cells was assessed using the standard ^51^Cr-release assay. Briefly, one million target cells were labeled at 37°C for 1 h with 100 μCi Na~2~^51^CrO~4~(New England Nuclear, Boston, MA). Target cells were washed and resuspended in TCM at 5 × 10^4^cells/ml. Five thousand target cells per well (100 μl) were added to a 96 well plate (Costar, Cambridge, MA) following the appropriate number of effector cells (100 μl/well). The defined effector:target (E:T) ratios were plated in triplicate. Cytotoxicity assays were performed at 37°C for 4 h. Percent specific lysis was calculated using the following equation: (ER - SR)/(MR - SR) × 100, where ER = experimental release, SR = spontaneous release and MR = maximum release. Granzyme B ELISPOT assay ------------------------ Granzyme B secretion was measured using the GrB ELISPOT assay as previously described \[[@B25]\]. Briefly, MultiScreen-IP plates (PVDF membrane, Millipore, Bedford, MA) were coated overnight at 4°C with 100 μl/well of anti-human GrB antibody (7.5 μg/ml in PBS, clone GB-10, PeliCluster, Cell Sciences, Norwood MA). Effector cells (100 μl/well) were added to triplicate wells at specified concentrations followed by 5 × 10^4^target cells per well (100 μl). After the specified effector-target cell incubation, the plates were washed and 100 μl/well of biotinylated anti-human GrB detecting antibody (0.25 μg/ml in PBS/1% BSA/0.05% Tween 20, clone GB-11, PeliCluster, Cell Sciences) was added. Plates were incubated for 3 h and 50 μl of Streptavidin-Alkaline Phosphatase (1:1500 in PBS/1% BSA, Gibco BRL Life Technologies) was added for 1 h. Spots were visualized with 100 μl/well of BCIP-NBT phosphatase substrate (KPL, Gaithersburg, MD) and subjected to automated evaluation using the ImmunoSpot Imaging Analyzer system (Cellular Technology Ltd, Cleveland, OH). IFN-γ ELISPOT assay ------------------- For assessing IFN-γ secretion, MultiScreen-IP plates (PVDF membranes, Millipore) were coated overnight at room temperature with 50 μl/well of anti-human IFN-γ antibody (20 μg/ml in PBS, Biosource, International, Camarillo CA) as previously described \[[@B25]\]. After effector and target cells were incubated at 37°C, the plates were washed and 50 μl/well of biotinylated anti-human IFN-γ antibody (1.3 μg/ml in PBS/1% BSA/0.05% Tween 20, BD PharMingen, San Jose, CA) was added. Plates were incubated with biotinylated antibody, washed and 50 μl of Streptavidin-Alkaline Phosphatase (1:1500 in PBS/1% BSA, Gibco BRL Life Technologies) was added. Spots were visualized and enumerated as described above. CD107a mobilization assay ------------------------- Degranulation of TALL-104 and LAK cells in response to target cell recognition was assessed by monitoring surface antigen expression of CD107a (lysosomal-associated membrane protein-1), a surface antigen transiently present on the cell surface after release of cytolytic granules. Expression of CD107a has been used as a marker to measure degranulation by flow cytometry \[[@B28],[@B29]\]. To distinguish target cells from effector cells, K562 cells were first labeled with PKH67 dye (Sigma) according to the manufacturer\'s instructions and washed extensively prior to use in the assay. Effector cells (2 × 10^5^) were resuspended in 400 μl of phenol red-free TCM in polystyrene tubes (Falcon, Franklin Lakes, NJ). PKH67-labeled K562 target cells (1 × 10^5^cells in 50 μl) were then added to appropriate tubes, the tubes were spun for 30 sec at 500 rpm and incubated for the specified periods in a 37°C water bath. Reactions were quenched with cold PBS and then 20 μl of CD107a-PE-Cy5 (PharMingen) was added. Mouse IgG~1~-biotin (Beckman-Coulter, Miami, FL) and streptavidin-PE-Cy5 (Caltag, Burlingame, CA) were utilized for controls. Tubes were incubated with antibody for 20 min at room temperature, washed with buffer and analyzed using a FACScan instrument (Becton-Dickinson Immunocytometry Systems, San Jose, CA) equipped with a 15 mW argon-ion laser. The percent of effector cells expressing CD107a was determined by gating the PKH67 negative (effector cells) population. Within the PKH67 negative gate, CD107a expression was determined from histograms based on forward and side scatter analysis. Effector cells run in the absence of target cells were evaluated for baseline expression of CD107a. The percentage of effector cells induced by target cells to express CD107a was calculated by subtracting background (spontaneous) expression from experimental samples. Statistical analysis -------------------- Statistical analysis was performed using Student\'s ttest and Pearson correlation coefficient (R^2^). Results ======= GrB release by stimulated TALL-104 cells ---------------------------------------- GrB ELISPOT assays were performed using a constant number of K562 targets and decreasing numbers of TALL-104 effector cells per well. As the number of TALL-104 cells decreased, the resulting number of GrB spots decreased linearly. Limited GrB spots were detected in wells containing TALL-104 alone demonstrating that the ELISPOT assay measured primarily stimulated release of GrB (Fig. [1](#F1){ref-type="fig"}). K562 alone did not secrete GrB (data not shown). The correlation between the number of spots per well and the number of TALL-104 was highly significant with a Pearson correlation coefficient of R^2^= 0.98. Therefore, the GrB ELISPOT assay is capable of measuring the frequency of effector cells stimulated to secrete GrB. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Granzyme B release by stimulated TALL-104 cells measured in the ELISPOT assay.TALL-104 cells (1 × 10^3^, 2 × 10^3^or 5 × 10^3^cells per well) were incubated with K562 targets (5 × 10^4^cells per well) for 4 h at 37°C in the Granzyme B ELISPOT assay (effector:target cell ratios are in brackets). Data are presented as average number of spots per well ± SD and are representative of 3 experiments with similar results. A significant correlation between the number of spots per well and effector cell number was observed (R^2^= 0.98). ::: ![](1479-5876-2-31-1) ::: Correlation of GrB secretion and cytotoxic activity --------------------------------------------------- In tandem with the GrB ELISPOT assay, TALL-104 cells were also utilized as effectors against K562 targets in the standard ^51^Cr-release assay to assess lytic activity. Significant specific lysis was observed at effector:target ratios from 10:1 to 1:1 (Table [1](#T1){ref-type="table"}). In contrast to the ELISPOT assay, that can measure the frequency of cytotoxic cells at effector:target ratios as low as 0.02:1, specific lysis was not significant in the ^51^Cr-release assay at cell ratios below 1:1 (Table [1](#T1){ref-type="table"}). However, when optimal ratios were used in each individual assay, similar trends between the number of GrB spots per well in the ELISPOT and specific lysis in the ^51^Cr-release assay were observed. These data demonstrate that the release of GrB is indicative of cytolytic activity. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Comparison of the GrB ELISPOT and ^51^Cr-release assays for measuring TALL-104 cell activity. ::: Effector:Target Ratio ^51^Cr-Release Assay^a^(% Cytotoxicity ± SE) GrB ELISPOT Assay^b^(Spots/well ± SE) ----------------------- ---------------------------------------------- --------------------------------------- 10:1 65 ± 4 TNTC^c^ 5:1 57 ± 10 TNTC 2.5:1 47 ± 11 TNTC 1:1 20 ± 5 TNTC 0.5:1 9 ± 2 TNTC 0.2:1 2 ± 1 316 ± 50 0.1:1 2 ± 1 144 ± 13 0.05:1 1 ± 1 80 ± 6 0.02:1 0 ± 1 33 ± 4 ^a^The ^51^Cr-release assay was performed with the effector:target cell ratio as specified for 4 h at 37°C. Data presented as average cytotoxicity ± SE (n = 3). ^b^The granzyme B ELISPOT assay was performed with K562 cells at a constant number of 5 × 10^4^cells per well and effector:target cell ratio as specified for 4 h at 37°C. Data presented as average spots/well ± SE (n = 3). ^c^TNTC= Too numerous to count accurately. ::: Secretion kinetics of GrB and IFN-γ ----------------------------------- To measure the secretion kinetics of GrB compared to IFN-γ, TALL-104 cells were incubated with K562 target cells for the indicated times in the ELISPOT assays. Secretion of GrB was compared to IFN-γ because IFN-γ is currently the standard utilized to measure the frequency of cytotoxic cells via the ELISPOT method. GrB secretion was observed within 10 min of effector-target cell interaction with optimal secretion around 3--4 h (Fig. [2](#F2){ref-type="fig"}). This observation is in contrast to maximum IFN-γ production, which was observed at 24 h of effector-target interaction (data not shown). These data are consistent with rapid effector cell degranulation (within minutes) upon contact with target cells \[[@B18],[@B19]\]. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Kinetics of Granzyme B and IFN-γ secretion by TALL-104 cells in the ELISPOT assays.TALL-104 cells (2.5 × 10^3^cells per well) were incubated with K562 cells (5 × 10^4^cells per well) in the Granzyme B and IFN-γ ELISPOT assays for 0.2, 0.5, 1, or 4 h at 37°C. Data are presented as average spots per well ± SD. Results are corrected for background and are representative of 3 experiments with similar results. ::: ![](1479-5876-2-31-2) ::: Mechanism of GrB secretion -------------------------- Effects of secretion inhibitors on the release of GrB and IFN-γ in the ELISPOT assays as well as expression of CD107a on effector cells were evaluated to determine the possible mechanism of GrB release measured in the ELISPOT assay. TALL-104 cells were treated with inhibitors of cellular secretion and assessed for their ability to release GrB and IFN-γ in the ELISPOT assays. BAPTA-AM chelates intracellular Ca^2+^and was utilized to block degranulation. Brefeldin A, which blocks protein secretion, was utilized to prevent secretion of newly synthesized proteins. Loading TALL-104 cells with BAPTA-AM resulted in a dose-dependent inhibition of GrB secretion (Fig. [3](#F3){ref-type="fig"}). Inhibition of GrB was not attributed to alterations in cell viability as TALL-104 cells loaded with as high as 100 μM of BAPTA-AM remained viable as assessed by trypan blue exclusion (data not shown). Brefeldin A treatment did not alter the secretion of GrB when TALL-104 cells were stimulated with K562 targets but did significantly (p\< 0.05) inhibit IFN-γ secretion (Fig. [4](#F4){ref-type="fig"}). Additionally, increased expression of the degranulation marker, CD107a, on TALL-104 cells was observed when TALL-104 cells were incubated with K562 target cells. Compared to TALL-104 cells alone, the addition of K562 targets induced 13.75% of TALL-104 cells to express CD107a (Fig. [5](#F5){ref-type="fig"}). Taken together, these data are consistent with synthesis of IFN-γ de novowhile GrB is released from preformed granules upon effector-target interaction. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Effect of BAPTA-AM on Granzyme B secretion by TALL-104 cells.TALL-104 cells were pretreated with the indicated concentrations of BAPTA-AM for 45 min at 37°C. TALL-104 cells (2.5 × 10^3^cells per well) were incubated with K562 (5.0 × 10^4^cells per well) for 4 h in the Granzyme B ELISPOT assay. Data are presented as spots per well ± SE (n = 3). Results are corrected for background. ::: ![](1479-5876-2-31-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Effect of Brefeldin A on Granzyme B and IFN-γ secretion in the ELISPOT assays.TALL-104 cells were preincubated with brefeldin A (5 μg/ml; 1 h at 37°C) prior to use in the ELISPOT assays. TALL-104 cells (2.5 × 10^3^cells per well) were incubated with K562 (5 × 10^4^cells per well) for 4 h in the Granzyme B and 24 h in the IFN-γ ELISPOT assays at 37°C. Data are presented as percent inhibition ± SE (n = 3). Results are corrected for background. ::: ![](1479-5876-2-31-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Stimulated TALL-104 cells express the degranulation marker CD107a.TALL-104 (2 × 10^5^cells) were incubated with K562 (1 × 10^5^cells) and surface expression of CD107a was determined by flow cytometric analysis. K562 were prelabeled with PKH67 dye to differentiate effector cells from targets. The histograms represent the PE-Cy5 CD107a positive cells determined by forward versus side scatter of the gated effector cells. TALL-104 cells alone are represented by the open histogram and TALL-104 cells incubated with K562 for 1 h are represented by the shaded histogram. The data is representative of 3 experiments with similar results. ::: ![](1479-5876-2-31-5) ::: Application of the GrB ELISPOT assay to assess innate immunity -------------------------------------------------------------- To address the potential clinical relevance of the GrB ELISPOT assay, LAK cells derived from PBMC, rather than the TALL-104 cell line, were utilized as effector cells. As shown in Figure [6](#F6){ref-type="fig"}, the number of GrB spots per well correlated with the number of LAK cells added, results comparable to those obtained with TALL-104 cells. Only wells containing both LAK and K562 target cells contained a substantial number of spots (Fig. [6](#F6){ref-type="fig"}). ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Granzyme B release by stimulated LAK cells measured in the ELISPOT assay.Human peripheral blood mononuclear cells (PBMC) were cultured with 100 U/ml of IL-2 for 5--6 days to induce LAK cells. LAK cells (1 × 10^3^, 3 × 10^3^or 1 × 10^4^cells per well) were incubated with K562 (5 × 10^4^cells per well) for 4 h at 37°C. Effector:target cell ratios are in brackets. Data are presented as average number of spots per well ± SD and are representative of 5 experiments with similar results. A significant correlation between the number of spots per well and effector cell number was observed (R^2^= 0.94). ::: ![](1479-5876-2-31-6) ::: The secretion kinetics of GrB and IFN-γ by LAK cells were similar to that observed for TALL-104 cells. GrB secretion was seen within 10 min of effector-target cell interaction with optimal secretion around 3 h and maximal IFN-γ production was observed at 24 h of effector-target interaction (Fig. [7](#F7){ref-type="fig"}). Similarly, stimulated secretion of GrB by LAK cells was associated with increased surface expression of the degranulation marker, CD107a. The correlation between the number of GrB spots and the number of cells expressing CD107a per 100,000 effector cells was significant (R^2^= 0.89, Fig. [8](#F8){ref-type="fig"}). These data demonstrate that the GrB ELISPOT assay accurately measures the rapid release of GrB by stimulated LAK cells and can enumerate GrB-secreting effector cells. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Kinetics of Granzyme B and IFN-γ secretion by LAK cells in the ELISPOT assays.Human peripheral blood mononuclear cells (PBMC) were cultured with 100 U/ml of IL-2 for 5--6 days to induce LAK cells. LAK cells (5 × 10^3^cells per well) were incubated with K562 cells (5 × 10^4^cells per well) in the Granzyme B and IFN-γ ELISPOT assays for 0.2, 0.5, 1, 4 or 24 h at 37°C. Data are presented as average spots per well ± SE (n = 6). Results are corrected for background. ::: ![](1479-5876-2-31-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Correlation between CD107a expression measured in the flow cytometric assay and Granzyme B release measured in the ELISPOT assay.Human peripheral blood mononuclear cells (PBMC) were cultured with 100 U/ml of IL-2 for 5--6 days to induce LAK cells. LAK cells (5 × 10^3^cells per well) were incubated with K562 cells (5 × 10^4^cells per well) in the Granzyme B ELISPOT and 2 × 10^5^LAK cells were incubated with 1 × 10^5^K562 in the CD107a mobilization assay for 0.5, 1 and 3 h. The ELISPOT data are presented as spots per 1.0 × 10^5^effectors ± SE (n = 9) and the CD107a data as the number of positive cells per 1.0 × 10^5^effectors ± SE (n = 9). All results are corrected for background. A significant correlation between the number of GrB spots and CD107a positive effector cells was observed (R^2^= 0.89). ::: ![](1479-5876-2-31-8) ::: LAK cells were utilized in the GrB ELISPOT, IFN-γ ELISPOT and ^51^Cr-release assays to evaluate the different assays for measuring innate immunity (Table [2](#T2){ref-type="table"}). Both ELISPOT assays were significantly more sensitive than the ^51^Cr-release assay. At effector:target ratios of 20:1-5:1, significant specific lysis can be measured by the ^51^Cr-release assay but the spots per well in the ELISPOT assays were too numerous to count accurately. Significant IFN-γ and GrB secretion was evident at effector:target ratios as low as 0.02:1 (1000 effectors/well) therefore, the optimal ratios for the ELISPOT assays are below the level of sensitivity of the ^51^Cr-release assay (Table [2](#T2){ref-type="table"}). However, when the optimal number of LAK cells are used in each individual assay, the amount of GrB and IFN-γ secreting cells in the ELISPOT assays and cytotoxicity in the ^51^Cr-release assay have shown cross-correlation with R^2^ values greater than 0.86 for all combinations. When comparing the two ELISPOT assays, the GrB ELISPOT is more rapid and may be a more direct measure of cytotoxic activity than the IFN-γ ELISPOT. Therefore, the GrB ELISPOT assay can elucidate the frequency of GrB secreting cells in primary LAK cultures and can be applied to measure innate immunity in clinically relevant samples. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of the GrB ELISPOT, IFN-γ ELISPOT and ^51^Cr-release assays for measuring Lymphokine Activated Killer (LAK) cell activity. ::: Effector:Target Ratio ^51^Cr-Release Assay^a^(% Cytotoxicity ± SE) IFN-γ ELISPOT Assay^b^(Spots/well ± SE) GrB ELISPOT Assay^c^(Spots/well ± SE) ----------------------- ---------------------------------------------- ----------------------------------------- --------------------------------------- 20:1 42 ± 10 TNTC^d^ TNTC 10:1 30 ± 10 TNTC TNTC 5:1 22 ± 6 TNTC TNTC 2.5:1 10 ± 3 TNTC TNTC 1:1 5 ± 1 TNTC TNTC 0.4:1 4 ± 1 227 ± 16 282 ± 18 0.2:1 2 ± 1 182 ± 8 193 ± 12 0.1:1 2 ± 2 121 ± 14 117 ± 12 0.04:1 1 ± 0 61 ± 13 54 ± 11 0.02:1 1 ± 0 29 ± 8 37± 7 ^a^The ^51^Cr-release assay was performed with the effector:target cell ratio as specified for 4 h at 37°C. Data presented as average cytotoxicity ± SE (n = 3). ^b^The IFN-γ ELISPOT assay was performed with K562 cells at a constant number of 5 × 10^4^cells per well and effector:target cell ratio as specified for 24 h at 37°C. Data presented as average spots/well ± SE (n = 3). ^c^The granzyme B ELISPOT assay was performed with K562 cells at a constant number of 5 × 10^4^cells per well and effector:target cell ratio as specified for 4 h at 37°C. Data presented as average spots/well ± SE (n = 3). ^d^TNTC= Too numerous to count accurately. ::: Discussion ========== NK cells are a key part of innate immunity due to their ability to secrete cytokines and mediate cytolytic activity and act as an important first line of defense against virally infected cells and tumor cells \[[@B30]\]. Although cytokine secretion by NK cells plays a role in regulating the adaptive immune response, cell-mediated cytotoxicity is the major effector function of NK cells. NK cells can mediate cytotoxicity by two main pathways, FasL-mediated and granule-mediated. The granule-mediated pathway, however, is dominant in NK cells with the release of GrB as one of the key factors in the lethal interaction that kills virus-infected and tumor cells \[[@B2],[@B5],[@B7],[@B8],[@B10],[@B15]-[@B19]\]. Therefore, evaluating the secretion of this molecule provides an indirect measure of cell-mediated cytotoxicity mediated via the release of granules. The GrB ELISPOT assay has been previously applied to enumerate the frequency of antigen-specific T cells and this release of GrB by T cells was indicative of cytolytic ability \[[@B24],[@B25],[@B31]\]. Moreover, the GrB ELISPOT assay requires significantly less effector cells than the standard ^51^Cr-release assay \[[@B24],[@B25]\]. In this study, we demonstrated that the GrB ELISPOT assay can be utilized to determine the frequency and potential cytolytic ability of innate immune cells. When TALL-104 or LAK cells were used as effectors, our results were in excellent agreement with results obtained when GrB-transfected CHO cells, T-cell lines or CTL stimulated in vitro, were employed as effector cells \[[@B24],[@B25]\]. TALL-104 or LAK cells secreted significant amounts of GrB only when stimulated with an NK-sensitive target cell line indicating that the GrB ELISPOT assay primarily detected stimulated, and not spontaneous (constitutive), release of GrB. Although assessing GrB release is an indirect measure of target cell lysis, a similar trend between the number of GrB spots per well in the ELISPOT assay and specific lysis in the ^51^Cr-release was observed. Therefore, the GrB ELISPOT assay is a viable alternative to the ^51^Cr-release assay for measuring MHC non-restricted cytolytic activity. Preformed granules that contain mature GrB are constitutively expressed in NK cells, thus NK cells are always armed with functional GrB. Preformed GrB is rapidly released upon recognition and conjunction of the effector cell and this process is Ca^2+^-dependent \[[@B18],[@B19],[@B32]\]. In contrast, IFN-γ is produced de novoupon activation and is secreted within hours \[[@B33]\]. The differences observed in the pattern of GrB and IFN-γ secretion suggest that the GrB measured in the ELISPOT assay is due to degranulation of pre-formed GrB whereas IFN-γ secretion results from protein synthesis de novo. To confirm that GrB measured in the ELISPOT assay was released via degranulation, effector cells were treated with inhibitors of cellular secretion: brefeldin A, BAPTA-AM and EGTA. Brefeldin A reversibly disrupts protein translocation from the endoplasmic reticulum to the Golgi apparatus, blocking the production and subsequent secretion of newly synthesized proteins \[[@B34]\]. BAPTA-AM and EGTA are both Ca^2+^-chelating agents and therefore can block Ca^2+^-dependent degranulation. Cell permeant BAPTA-AM contains acetoxymethyl ester groups that are removed by cytosolic nonspecific esterases, trapping the chelator within the cell. In a series of experiments, treatment of effector cells with brefeldin A significantly decreased IFN-γ, but not GrB secretion, whereas, BAPTA-AM abrogated GrB secretion but its effect on IFN-γ secretion was not apparent (data not shown). EGTA, which sequesters extracellular Ca^2+^, could also abrogate GrB secretion measured in the ELISPOT assay. However, significantly higher concentrations of EGTA (4--40 mM) than BAPTA-AM were needed to block GrB secretion (data not shown). These data are in good agreement with recently published findings \[[@B35]\] which demonstrate that higher concentrations of EGTA are needed to block degranulation compared to BAPTA-AM. Although adherence of cytolytic cells to their targets is a prerequisite for granule release, this interaction is dependent on Mg^2+^, but not Ca^2+^\[[@B19]\]. Thus, the decrease in GrB secretion after BAPTA-AM and EGTA treatment cannot be attributed to inhibition of adhesion between the effector and target cells, but results from inhibition of degranulation. Recently, a flow cytometric assay based on CD107a (lysosomal-associated membrane protein-1) mobilization was developed to measure degranulation of cytolytic cells \[[@B28],[@B29]\]. CD107a is a vesicle membrane protein of cytolytic granules that is transiently expressed on the surface of effector cells during degranulation. Correlations between direct lytic ability and surface expression of CD107a on effector cells has been shown, indicating that CD107a expression is a reliable measure of cytolytic capacity \[[@B28],[@B29]\]. In our study, CD107a expression correlated with GrB secretion (R^2^= 0.89). Both the release of GrB and the expression of CD107a was observed as early as 30 min after LAK cells were stimulated with relevant target cells, and increased with extended effector-target cell interaction. These data confirm that the GrB ELISPOT assay is an excellent measure of cytotoxic capacity mediated by effector cell degranulation. When the ELISPOT assays were directly compared to the ^51^Cr-release assay, they demonstrated higher sensitivity with both TALL-104 and LAK cells as effector cells, data consistent with previous studies \[[@B24],[@B25],[@B36],[@B37]\]. The ELISPOT assays enumerate antigen specific lymphocyte frequency by measuring secretion of specific immune proteins engaged in the specific pathway utilized to mediate lysis of target cells, whereas the ^51^Cr-release assay measures cytotoxic ability regardless of the mechanisms of the killing. Therefore, unlike the ^51^Cr-release assay, the ELISPOT assays are both qualitative and quantitative. It is important to emphasize that ^51^Cr-release and the GrB ELISPOT assay measure different aspects of cell-mediated killing -- target cell death and effector cell function, respectively. A limitation of the GrB ELISPOT assay is that it measures degranulation, not direct target cell lysis. As such, degranulation may not always equate to cell death if target cells contain serpin proteinase inhibitor 9 (PI9), a protein that inhibits the proteolytic activity of GrB, \[[@B38]\] or if effector cells are perforin deficient. The GrB ELISPOT assay also does not account for cytotoxicity mediated by FasL pathway. Therefore, when appropriate, the two assays should be used in concert. However, the high sensitivity and specificity of the ELISPOT assay are beneficial for monitoring clinical trials where frequently there are limited numbers of patients\' cells available or target cells cannot be effectively labeled. Target cells that resist or do not tolerate labeling, such as some primary tumor cells, can still be utilized in the GrB ELISPOT assay. Additionally, compared to the ^51^Cr-release assay, the GrB ELISPOT provides the precusory frequency of cells with the potential to kill targets. Although a number of flow cytometric assays have been developed to assess target cell cytotoxicity as well as identify the phenotype of the effector cells mediating the immune response, these assays are not as sensitive as the ELISPOT assay. The use of the IFN-γ ELISPOT assay as a surrogate measure for CTL and NK responses has recently gained increased application as an alternative to the ^51^Cr-release assay \[[@B36],[@B39]-[@B43]\]. However, the IFN-γ ELISPOT assay may not be an accurate measure of cytotoxic lymphocytes since 1) non-cytotoxic cells can secrete IFN-γ and 2) CTL with lytic activity do not always secrete IFN-γ \[[@B44]-[@B48]\]. The release of GrB is a more specific measure of cytotoxic lymphocytes than IFN-γ because the expression of GrB is restricted to CTL and NK cells \[[@B8],[@B49],[@B50]\]. Therefore, the GrB ELISPOT assay may be a more direct measure of innate immunity compared to the IFN-γ ELISPOT. The high sensitivity and specificity of the ELISPOT assays are beneficial for monitoring clinical trials where frequently there are limited numbers of patients\' cells available. As such, simultaneous use of the IFN-γ and GrB ELISPOT assays may provide important immunological insight into patient responses that may then be directly assessed against clinical outcome. Conclusion ========== Our data demonstrate that the GrB ELISPOT assay is a viable alternative to the standard ^51^Cr-release assay to measure granule-mediated cytotoxicity and could be easily adapted to reliably and accurately measure MHC non-restricted cytotoxicity indicative of NK and LAK activity. The GrB ELISPOT assay measured GrB release due to degranulation of stimulated effector cells. Additionally, GrB is a more specific candidate marker than IFN-γ to measure the cytotoxic capacity of innate immune effector cells such as NK cells. Competing interests =================== None declared. Author\'s contributions ======================= KSW and AM designed the study, analyzed the data and drafted the manuscript. KSW carried and prepared the cell lines and performed the majority of the immunogical assays. MWB performed and analyzed the flow cytometric CD107a assay data. SS participated in the design of the study and preparation of the manuscript. TS, DK and MB participated in the design of the study and served as expert advisors. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by the National Cancer Institute, SAIC-Frederick, Inc., Contract N01-C0-12400. The content of this publication does not necessarily reflect the view or the policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.	PubMed Central	2024-06-05T03:55:48.451530	2004-9-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522821/", "journal": "J Transl Med. 2004 Sep 20; 2:31", "authors": [ { "first": "Kimberly A", "last": "Shafer-Weaver" }, { "first": "Thomas", "last": "Sayers" }, { "first": "Douglas B", "last": "Kuhns" }, { "first": "Susan L", "last": "Strobl" }, { "first": "Mark W", "last": "Burkett" }, { "first": "Michael", "last": "Baseler" }, { "first": "Anatoli", "last": "Malyguine" } ] }
PMC522822	Background ========== A cholesterol lowering effect has been reported for a variety of soluble dietary fibres \[[@B1]-[@B5]\]. In February, 1998, the U.S. Food and Drug Administration authorized the use, on food labels and food labelling, of health claims on the association between soluble fibre from psyllium seed husk and a reduced risk of coronary heart disease \[[@B6]\]. Among the suggested mechanisms by which soluble fibre lowers cholesterol is the binding of bile acids in the intestinal lumen resulting in decreased absorption and increased faecal excretion of them \[[@B7]-[@B15]\]. The ensuing bile acid depletion increases hepatic demand for the de novo synthesis of bile acids from cholesterol. This requirement is met, in part, by increased hepatic LDL receptor activity, which in turn reduces circulating LDL. The accumulation and concentration of bile in the gallbladder is a continuous process. In rats which lack a gallbladder, the biliary excretion rate of bile salts is maximal at night \[[@B16]\] and bile is stored in the gallbladder during an overnight fast. To the extent that the cholesterol-lowering effect of psyllium requires an interaction with bile, the magnitude of its cholesterol-lowering effect should vary with the quantity of bile in contact with a given amount of psyllium. The gallbladder empties with a meal and the total quantity of bile salt presented to the small bowel should be highest when this emptying occurs at breakfast, following a fast and the associated overnight accumulation of secreted bile. Conversely, psyllium presented to the gut following a short fast and without the full stimulus to gallbladder emptying of an associated meal should result in an encounter with a smaller amount of bile. We accordingly set out to test the hypothesis that administering psyllium with breakfast would have a significantly greater cholesterol-lowering effect than would taking a similar dose at bedtime. Since the predominance of the literature with regard to the cholesterol-lowering effect of psyllium is in individuals with hypercholesterolemia, a secondary goal of our study was to confirm the cholesterol-lowering effect of psyllium in subjects with \"average\" cholesterol levels. Methods ======= Patients identified in gastroenterology practices as requiring long-term treatment with psyllium, typically for chronic constipation or the irritable bowel syndrome, were invited to participate in the study. In addition, individuals who had received dietary counselling in a lipid clinic regarding cholesterol-lowering and who subsequently had cholesterol levels deemed not to require further intervention because they met targets set out in clinical practice guidelines were invited by clinic staff to participate. Subjects were deemed ineligible if they were under age 18 years, were under active treatment for hyperlipidemia, had total cholesterol greater than 7.00 mmol/L, required alterations in dosage of medications which might have an effect on lipid levels, had had a gastrectomy, had any disease which is associated with hyperlipidemia, were receiving a bile acid binding resin, or if they did not eat breakfast regularly. The study was approved by the Conjoint Health Research Ethics Board of the University of Calgary, Faculty of Medicine. Subjects were given a description of the study indicating our interest in comparing the relative efficacy of hs versus am dosing with psyllium without indicating the specific hypothesis, and were asked to sign a consent form. Gastroenterology patients were given a high fibre diet sheet as part of their therapeutic regimen and were asked to take this on a continuing basis, beginning one month prior to the initiation of the study. The diet sheet emphasized dietary sources containing predominantly insoluble fibre. Lipid clinic patients had all received in-depth counselling regarding dietary measures for hypercholesterolemia, including a high fibre regimen, and had implemented their dietary changes at least one month before beginning the study. An unsweetened psyllium preparation, \"Novo-Mucilax\" \[NovoPharm\], providing three grams of hydrophilic mucilloid per 6.2 gram powder, and a scoop known to provide at least ten grams of psyllium were provided. Containers were numbered and weighed at the conclusion of each test interval. Subjects were randomized to initially take a scoop full of psyllium either with breakfast or at bedtime. Using a crossover design, psyllium was taken in the morning or evening for eight weeks and at the alternate time for the subsequent eight weeks. Determinations of serum total cholesterol, LDL, HDL and triglycerides were made before beginning psyllium, at eight weeks and at sixteen weeks after commencing its use. A trained dietician obtained a dietary history and patients were weighed at the beginning and at the conclusion of the study. All lipid determinations were undertaken following a 14 hour fast and analyses were done in a central laboratory. After initial data inspection based on boxplots and summary measures, cholesterol values at 8 and 16 weeks were examined using analysis of variance, taking into account treatment, period, and between and within subject effects in accordance with the cross-over design. The pattern of change from baseline to 16 weeks was evaluated using paired t-tests. Results ======= Of 86 subjects beginning the study, 33 of those referred from the gastroenterology clinics and 30 of those referred from the lipid clinic completed it. Of those withdrawing, eight did so because they could not tolerate the psyllium or it was felt to interfere with prescribed medication, 2 had elevated lipids, 4 did not complete all the required blood work, 5 were unable to comply with the protocol because of work or lifestyle changes and 4 developed intercurrent diseases which precluded completing the protocol. The age range of subjects was 18 to 77 years, mean 53 +/- 13 years, including 16 men and 47 women. The mean dose of psyllium taken was 12.7 +/ - 2.3 g for morning dosing and 12.7 +/- 2.2 g when taken in the evenings. Values for total, LDL, HDL cholesterol and triglycerides at baseline, eight and sixteen weeks for various subgroupings are tabulated in tables [1](#T1){ref-type="table"}, [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}, respectively. Data for changes from baseline for total, LDL and HDL cholesterol and triglycerides are listed in table [5](#T5){ref-type="table"}. Changes from baseline were not significant for any parameter. Furthermore, the confidence intervals given in Table [5](#T5){ref-type="table"} show that this data essentially excludes a clinically significant psyllium effect. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Total Cholesterol in mmol/L \[SD\] ::: \[n\] Baseline 8 wks 16 wks ------------------------------ --------------- --------------- --------------- AM first 5.76 \[0.94\] 5.77 \[1.01\] 5.80 \[0.95\] PM first 5.47 \[0.98\] 5.61 \[1.10\] 5.57 \[1.06\] Females \[47\] 5.72 \[0.96\] 5.79 \[1.07\] 5.74 \[0.99\] Males \[16\] 5.29 \[0.94\] 5.39 \[0.96\] 5.48 \[1.04\] GI subjects \[33\] 5.00 \[0.83\] 5.06 \[0.91\] 5.07 \[0.84\] Lipid clinic Subjects \[30\] 6.28 \[0.59\] 6.39 \[0.69\] 6.35 \[0.70\] All subjects \[63\] 5.61 \[0.96\] 5.69 \[1.05\] 5.68 \[1.00\] \"AM first\" are subjects taking psyllium in the morning for the first 8 weeks and in the evening for the second 8 weeks. Vice versa for \"PM first\". ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### LDL CHOLESTEROL in mmol/L \[SD\] ::: Baseline 8 wks 16 wks ----------------------- --------------- --------------- --------------- AM first 3.55 \[1.01\] 3.51 \[1.04\] 3.52 \[0.94\] PM first 3.39 \[0.90\] 3.43 \[1.03\] 3.45 \[0.92\] Females 3.55 \[0.92\] 3.54 \[1.04\] 3.52 \[0.92\] Males 3.22 \[1.03\] 3.24 \[0.99\] 3.39 \[0.97\] GI subjects 2.89 \[0.81\] 2.85 \[0.85\] 2.91 \[0.72\] Lipid clinic subjects 4.09 \[0.66\] 4.15 \[0.72\] 4.11 \[0.69\] All subjects 3.47 \[0.95\] 3.47 \[1.03\] 3.48 \[0.92\] ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### HDL CHOLESTEROL in mmol/L \[SD\] ::: Baseline 8 wks 16 wks ----------------------- --------------- --------------- --------------- AM first 1.38 \[0.29\] 1.42 \[0.37\] 1.42 \[0.33\] PM first 1.38 \[0.40\] 1.39 \[0.36\] 1.36 \[0.33\] Females 1.41 \[0.30\] 1.44 \[0.30\] 1.42 \[0.29\] Males 1.28 \[0.45\] 1.29 \[0.50\] 1.30 \[0.43\] GI subjects 1.36 \[0.39\] 1.41 \[0.43\] 1.37 \[0.38\] Lipid clinic subjects 1.40 \[0.30\] 1.40 \[0.28\] 1.41 \[0.27\] All subjects 1.38 \[0.35\] 1.41 \[0.36\] 1.39 \[0.33\] ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### TRIGLYCERIDES in mmol/L \[SD\] ::: Baseline 8 wks 16 wks ----------------------- --------------- --------------- --------------- AM first 1.81 \[0.70\] 1.84 \[0.86\] 1.86 \[0.91\] PM first 1.53 \[0.52\] 1.69 \[0.70\] 1.65 \[0.61\] Females 1.66 \[0.61\] 1.72 \[0.70\] 1.76 \[0.71 Males 1.70 \[0.68\] 1.89 \[0.99\] 1.72 \[0.94\] GI subjects 1.63 \[0.54\] 1.71 \[0.82\] 1.70 \[0.72\] Lipid clinic subjects 1.71 \[0.71\] 1.82 \[0.74\] 1.80 \[0.83\] All subjects 1.67 \[0.35\] 1.76 \[0.78\] 1.75 \[0.77\] ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Mean changes from baseline at 8 and 16 weeks \[mmol/L\] and 95% confidence intervals \[all patients, n = 63\] ::: 8 wks 16 wks --------------- ----------------- ---------- ------------------ ---------- TOTAL 0.080 p = 0.26 0.068 p = 0.28 CHOLESTEROL \[-0.061,0.22\] \[-0.056,0.19\] LDL 0.002 p = 0.98 0.018 p = 0.75 CHOLESTEROL \[-0.13,0.14\] \[-0.098,0.13\] HDL 0.028 p = 0.28 0.012 p = 0.61 CHOLESTEROL \[-0.023,0.08\] \[-0.035,0.059\] TRIGLYCERIDES 0.094 p = 0.22 0.082 p = 0.22 \[-0.059,0.25\] \[-0.050,0.21\] ::: The mean caloric intake or the intakes of fat or fibre did not change significantly during the study \[table [6](#T6){ref-type="table"}\]. There was a small but significant increase in the weights of subjects, which was felt to reflect the high proportion of subjects beginning the study in the fall and the associated reduction of physical activity during the subsequent winter months. There was no apparent relationship between change in cholesterol and change in weight \[correlation = .1, p = .15\]. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Weight and Dietary Intake \[SD\] at Baseline and at 16 Weeks ::: Baseline 16 wks ----------------- --------------- --------------- ------------ WEIGHT \[Kg\] 74.6 \[16.5\] 75.4 \[16.9\] p = 0.0039 ENERGY \[kcal\] 1649 \[409\] 1689 \[419\] p = 0.29 FAT \[g\] 49.1 \[20.3\] 49.8 \[20.2\] p = 0.66 FIBER \[g\] 19.7 \[7.96\] 19.7 \[7.28\] p = 0.99 ::: Discussion ========== We failed to prove our hypothesis that administration of psyllium in the morning would have a greater cholesterol-lowering effect than it would in the evening. Not only was there no observable difference in lipid levels between the crossover periods but the daily ingestion of a greater daily dose than the 10.2 g of psyllium for which the FDA allows health claims to be made \[[@B6]\] had no effect on lipid levels in our study group. No change in any lipid parameter, including total and LDL cholesterol was observed. No difference was found when subgroup analysis was undertaken for the sex of the patients, the time of day they took their psyllium, or whether they were recruited from the gastroenterology clinic or the lipid clinic. We used a crossover design since this was the most appropriate one for the primary question being addressed; accordingly, our study did not include a control group. However, the nature of the study should not have provided any motivation for study subjects to adopt any new lifestyle or dietary changes beyond those implemented well before the introduction of psyllium. Observational data has been shown to provide valid information, which is consistent with that observed in randomized, controlled trials \[[@B17],[@B18]\]. The nature of the intervention was in keeping with those undertaken in day-to-day clinical practice and the protocol used should, therefore, have high \"clinical relevance\". Failure of lipid-lowering by psyllium has also been demonstrated in twenty hypercholesterolemic children \[[@B19]\], in twenty-four hyperlipidemic adults \[[@B20]\] and in a large observational study of elderly patients taking psyllium \[[@B21]\]. A report of lipid-lowering therapies in hypercholesterolemic veterans showed only a 2% reduction in LDL cholesterol and a small increase in LDL/HDL ratio in patients taking psyllium, but does not provide a measure of statistical significance \[[@B22]\]. One study revealed no difference in total cholesterol-lowering compared to placebo, but a reduction of LDL cholesterol resulted from psyllium treatment \[[@B23]\]. Another demonstrated no difference in total cholesterol-lowering compared to placebo, a reduction of LDL in 11 \"responders\" and no change in 9 \"nonresponders\" \[[@B24]\]. A reduction of HDL cholesterol has been noted in some studies \[[@B25]-[@B27]\] and was associated with changes in LDL/HDL ratios similar to placebo treatment \[[@B25],[@B26]\]. Published studies include few normocholesterolemic subjects. Cholesterol reduction was observed in 7 normal men \[[@B28]\] and in 5 of 9 subjects \[[@B29]\], in both studies after 3 weeks of treatment. A reduction of cholesterol levels was also observed in 12 elderly patients given psyllium for 4 months \[[@B30]\], while 5 normocholesterolemic subjects in another study showed no reduction after 2 to 7 months of treatment \[[@B31]\]. A meta-analysis of 17 studies of patients with hypercholesterolemia has suggested a small but significant cholesterol-lowering effect of psyllium \[[@B2]\]. All of these investigations were associated in one way or another with the product manufacturer. Additional studies have also indicated some cholesterol-lowering by psyllium in hypercholesterolemic individuals \[[@B32]-[@B37]\] or in diabetics \[[@B38]-[@B40]\]; however, much of this work is uncontrolled and some protocols have specifically excluded premenopausal women \[[@B33],[@B38]\]. The association of cholesterol-lowering effects with psyllium may be weakened in some studies by the use of a supplement containing additional forms of soluble fibre \[[@B42]\] or by apparent differences in intake of calories \[[@B43]-[@B46]\], soluble fibre \[[@B25]\] or cholesterol \[[@B47]\] in control and treatment groups or periods. Several reports include only small numbers of patients and/or are of short duration. There is a strong predominance of male subjects in these publications and some protocols incorporate additional treatment interventions \[[@B20],[@B48]\]. Several factors may contribute to the difference between our observations and those of others. A meta-analysis has demonstrated that the initial level of cholesterol was highly predictive of the subsequent reduction of cholesterol by oat bran \[[@B49]\]. A greater effect of psyllium in men compared to women has been suggested \[[@B23],[@B46]\] and a diet high in soluble fibre produced less cholesterol-lowering in post menopausal women than in men \[[@B10]\]. Soluble fibre has a lesser effect on lipid metabolism in female than in male guinea pigs \[[@B50]\] and there is a sex-based difference in mechanism of action in this animal \[[@B51]\]. Oat bran fails to lower cholesterol in young women, in contrast to men and older women \[[@B52]\]. The dominance of women in our study, the \"normal\", or only slightly abnormal cholesterol states of our subjects and the relatively young ages of some of them may, accordingly, account for some of the variance of our observations with some of those previously reported. The small increase in the weight of subjects is believed to be have resulted from reduced physical activity. In a meta-analysis of the effect of weight reduction on lipids, predominantly through dietary change, a reduction in total cholesterol of 0.05 mmol/L and of 0.02 mmol/L in LDL cholesterol per kilogram of weight lost was identified \[[@B53]\]. Dietary intakes were stable throughout our study and the average weight gain of less than one kilogram is very unlikely to have raised cholesterol levels to a degree sufficient to offset a significant cholesterol lowering effect of psyllium. A small cholesterol-lowering effect of psyllium appears to occur in hypercholesterolemic individuals, at least in men and possibly postmenopausal women. The notion of a benefit accruing to the general population requires additional study. The promotion of foods containing psyllium as reducing the risk of heart disease for the population at large \[[@B6]\] may be premature. Additional study is required and this should be undertaken in a manner that is free from concern regarding the possibility of publication bias which Brown L, Rosner B, Willett WW and Sacks FM have raised \[[@B2]\]. Conclusion ========== The timing of psyllium administration had no effect on cholesterol-lowering and, in fact, no cholesterol-lowering was observed. Conclusions regarding the effectiveness of psyllium for the prevention of heart disease in the population at large may be premature. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= GVR carried out the study design, data review and writing. EAS carried out the study design and data review. RB carried out the study design, data review and statistical analysis. ALE carried out the study design and data review. Acknowledgements ================ We are indebted to Gwendolyn Shah and Sandra Sharkey for their assistance with the interviews, instruction of subjects and with the gathering of the data and to Victoria Stagg for her help with the statistical analysis.	PubMed Central	2024-06-05T03:55:48.454159	2004-9-28	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522822/", "journal": "Nutr J. 2004 Sep 28; 3:17", "authors": [ { "first": "Guido MA", "last": "Van Rosendaal" }, { "first": "Eldon A", "last": "Shaffer" }, { "first": "Alun L", "last": "Edwards" }, { "first": "Rollin", "last": "Brant" } ] }
PMC522823	Background ========== Buruli ulcer (BU) is a skin disease, which usually begins as a painless nodule or papule and may progress to massive skin ulceration. If untreated BU may lead to extensive soft tissue destruction, with inflammation extending to deep fascia. The parts of the body most affected are the extremities. Subsequent complications may include contractural deformities. The main form of treatment is wide excisional surgery, including amputation of limbs, which requires prolonged hospitalization and is thus a significant burden on hospital resources and budgets. In recent years, there has been increased incidence of BU in West Africa (including Benin, Burkina Faso, Cote d\'Ivoire, Ghana, Guinea, Liberia and Togo), Mexico, French Guyana, Papua New Guinea and Australia. The disease seems to affect mostly impoverished inhabitants in remote and rural areas; children are the most vulnerable, accounting for about 70% of the cases \[[@B1]\]. The World Health Organization (WHO) has recognized BU as the third most prevalent mycobacterial disease after tuberculosis and leprosy and has called for urgent action to control it \[[@B2]\]. The causative agent of BU is Mycobacterium ulcerans(MU), which was first described in Bainsdale, Australia, in 1948 \[[@B3]\]. From the medical point of view, MU is among the group of mycobacteria that are potentially pathogenic in humans and animals under special circumstances \[[@B4]\]. It is suggested MU enters through a small break or trauma in the skin because it is not known to penetrate through intact or healthy skin \[[@B5],[@B6]\]. Portaels et al. \[[@B7]\] have suggested that insects may be involved in the transmission of the disease because insects found in the roots of trees tested positive with the mycobacterium. Marsolliers et al. \[[@B8]\] found through an experimental study that the bite of MU-infected waterborne insects transmitted infection to mice. In terms of human infection, however, the reservoir of MU and the mode of transmission of BU are still unclear \[[@B4],[@B9]-[@B12]\]. Epidemiological data suggest that environmental factors such as climate, soil, geology, geochemistry, etc. may indirectly influence or contribute to MU infection \[[@B4]\]. In addition, the frequencies of some diseases caused by mycobacteria indicate that species are distributed geographically \[[@B13]\]. For example, MU has been observed mainly in the tropics and \[[@B4]\] especially in anthropogenically-polluted areas \[[@B14]\]. Since MU is known to be present in nature although its reservoir is not known and since the epidemiology of BU is still unclear, there is a need to have a better understanding of environmental, ecological, and behavioural factors that predispose to infection. Spatial analysis potentially contributes important information leading to the understanding of the epidemiology and etiology of BU. The main objective of this paper is to explore relationships between some spatial environmental factors and the prevalence of BU. Results ======= For the buffers tested, Pvalues range from 0.09 to 0.46. The buffers with the highest Pvalues (i.e., 0.09) are 100 m for drainage channels and 400 m for farmlands. With these buffers 24 of the 61 settlements (i.e., 39%) fall within 100 m of arsenic-enriched drainage channels (Figure [4](#F4){ref-type="fig"}) and 41 of the 61 settlements (i.e., 67%) fall within 400 m of As-enriched farmlands (Figure [5](#F5){ref-type="fig"}). The mean BU prevalence within the drainage buffer is 0.7% whereas the mean BU prevalence inside the farmland buffer is 0.6%. Thus the naturally smaller number of settlements within the drainage buffer (i.e., 24) has a slightly higher BU prevalence than the relatively larger number of settlements within the farmland buffer (i.e., 41). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Map of As-enriched farmlands.Distances to As-enriched farmlands (red; with \> 15 ppm As in stream sediments) and locations of villages with BU cases. ::: ![](1476-072X-3-19-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Map of As-enriched streams.Distances to As-enriched streams (black; with \> 15 ppm As in stream sediments) and locations of villages with BU cases. ::: ![](1476-072X-3-19-5) ::: Discussion ========== Siting of rural settlements in the study area is based primarily on proximity to and availability of water for drinking and other domestic purposes. Consequently, many settlements are located within the optimum buffer distance of 100 m from drainage channels. Where water is abstracted from drainage channels enriched in arsenic, chronic ingestion of arsenic-enriched water through drinking and cooking is likely. This renders the inhabitants susceptible to several kinds of diseases \[[@B15],[@B16]\] including BU. Amofah et al. \[[@B17]\] studied 90 BU patients and found that 52 used surface water as the source of their drinking water. The result of the statistical analysis corroborates this observation in that BU prevalence is highest where the inhabitants have ready access to domestic water supplies from arsenic-enriched surface drainage. Subsistence farmlands, especially those that depend partially on irrigation, tend to be located along stream floodplains. Soils in these floodplains have a high cation exchange capacity \[[@B18]\] so that, where streams carry high concentration of arsenic, there is accumulation of arsenic in the soils of the floodplains. These high concentrations of arsenic are in part taken up by the foodcrops grown there \[[@B19]-[@B21]\]. The results of the statistical analysis suggest that a high proportion of settlements with high BU prevalence exploit such floodplain farmlands enriched in arsenic. Through consumption of arsenic-enriched drinking water and arsenic-enriched foodcrops, inhabitants in some settlements in the Amansie West District are prone to chronic ingestion of higher-than-average (but sub-toxic) levels of arsenic. Arsenic interacts with and inhibits several enzymes in the body \[[@B15]\] leading to several multisystemic non-cancer effects \[[@B16]\], which could predispose to defect the immune system \[[@B22]\]. Subjects exposed to high levels of arsenic concentrations were known to have impaired immune response \[[@B23]\]. Immunosuppression due to arsenic has been found to defect antigen processing of splenic macrophages with consequent defective mechanism of helper T-cells \[[@B24],[@B25]\]. Down-regulation of the immune system is known to be a risk factor for the development of BU \[[@B26],[@B27]\]. Several studies \[e.g., \[[@B28]-[@B31]\]\] have reported of impaired resistance to viral/bacterial infection via arsenic ingestion. Conclusions =========== The results of this study reveal spatial dependency of BU prevalence upon proximity to drainage channels and farmlands containing \> 15 ppm arsenic. Proximity implies chronic exposure to and/or ingestion of elevated concentrations of arsenic, which influences susceptibility to infection. Methods ======= Research hypotheses ------------------- It has been consistently theorized that BU is acquired when MU enters the body through a skin rupture \[[@B32],[@B27]\]. However, several people who were affected by the disease do not recall having any break or trauma in their skin prior to being infected \[[@B33]\]. A possible alternative is entry through non-ruptured but unusually unhealthy or thin skin. Several dermatological diseases (e.g., Bowen\'s disease, hyperkeratosis, hyperpigmentation) are related to arsenic ingestion and exposure \[[@B34]\]. Bioaccumulation of arsenic in the fatty tissues of the skin \[[@B35]\], due to its high lipid solubility \[[@B36],[@B37]\], may provide a favourable environment for MU in the skin because arsenic is known to help microorganisms grow \[[@B38]\]. It can be hypothesized, therefore, that (a) arsenic induces MU adhesion to human tissues and (b) arsenic influences the ability of MU to establish BU. In a case study, Amofah et al. \[[@B17]\] reported that about 44% of the BU patients were farmers whilst about 54% were school children. In Ghana many children help their parents on farms. Not only do farmers and children come in contact with natural drainage areas on their journeys to and from their farmlands, but also the farms are located near water bodies or drainage systems for obvious irrigation purposes \[[@B39]\]. If farmlands and surface drainage channels are contributory factors to BU, farmlands and surface drainage channels enriched in arsenic may contribute to still higher prevalence of BU. Research methodology -------------------- Spatial analysis of data provides opportunities for epidemiologists to study associations between environmental factors and spatial distribution of diseases \[[@B40]\]. A geographic information system (GIS) is capable of analyzing and integrating large quantities of geographically distributed data as well as linking geographic data to non-geographic data to generate information useful in further scientific (or medical) research and in decision-making. In this study, topographic map data, stream sediment geochemical data for arsenic, ASTER satellite imagery and locations of settlements with BU cases were the basic data inputs into the GIS. Spatial data processing was carried out (a) to delineate arsenic-enriched catchment basins based on arsenic concentrations in stream sediment samples, (b) to delineate farmlands from ASTER satellite imagery and determine arsenic-enriched farmlands based on catchment basin data and (c) to extract drainage channels from the topographic map and determine arsenic-enriched drainage channels based on arsenic-enriched catchment basins. Proximity analysis was undertaken to determine spatial relationships between BU-affected areas and the arsenic-enriched areas determined from the data inputs. The study area -------------- ### History of BU in Ghana The study area is in Ghana, where the first case of BU was reported in 1971 and, between 1991 and 1997, more than 2000 cases have been reported \[[@B41]\]. The disease has affected all of the ten regions and at least 90 of the 110 districts in Ghana \[[@B42]\]. The Ashanti Region is the worst affected, accounting for about 60% of all reported cases, of which the greatest percentage is in the Amansie West District (Figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The study area.Amansie West District, Ghana, showing the study area (box) and villages with BU cases (black dots). ::: ![](1476-072X-3-19-1) ::: ### Location of study area The Amansie West District lies between latitudes 6°N and 6°45\'N and longitudes 1°30\'W and 2°15\'W. It covers an area of about 1,136 km^2^. The district capital, Manso Nkwanta, is about 40 km south of Kumasi. The district is drained by the Offin and Oda rivers. Vegetation in the district is composed mainly of secondary forests, thicket, forb regrowth (i.e., soft-stemmed leafy herbs, mostly the weeds, which appear on farms and have to be cut regularly) and swamp vegetation. Vegetation thrives in ferric fluvisols, which are the major soil types in the district. These soils have been developed through yearly rainfall ranging from 125 to 200 cm with temperatures of 22 to 30°C. The landscape of the district varies from gentle to broken. The district is underlain by Lower Proterozoic Birimian and, to a lesser extent, Tarkwaian rocks. Throughout Ghana, Birimian rocks of West Africa are mainly volcanic greenstones with intervening sedimentary rocks and granitoid intrusions, in places containing deposits composed of pyrite, arsenopyrite, minor chalcopyrite, sphalerite, galena, native gold and secondary hematite \[[@B43]\]. The district has about 310 settlements (though not all these settlements are mapped) with a population in 2000 of 108,726. There are approximately equal percentages of males and females (49% and 51%, respectively), of whom 70% are farmers and 22% are engaged in legal and \'galamsey\' (or illegal) mining. The study area is the east-central part of the Amansie West District (covered by a single topographic map sheet, 0602C1), with an area of about 623 km^2^, comprising 61 settlements and including the Bilpraw goldmine (formerly a treasure mine of the Ashanti Kings). The BU cases per settlement range from 1--29. ### Materials The following are the sources of spatial data input to the GIS. • Incidence of BU per settlement in 1999, obtained from Korle-BU Teaching Hospital, Accra, Ghana. • Settlement population estimates for 2000, projected by the Ministry of local government and rural development. • Topographic map (Sheet 0602C1, 1974, at a scale of 1: 50,000), a single sheet covering the study area, obtained from the Survey Department, Accra, Ghana. • Location map (at scale of 1: 62,500) of stream sediment samples collected in part of the Amansie West District in 1992 and list of arsenic concentrations determined in these samples, obtained from the Geological Survey Department, Accra, Ghana. • Boundary map (at scale of 1: 250,000, surveyed in 1991) of the district, obtained from the Amansie West District Administration. • ASTER imagery (level 1B) acquired on 15/01/2002, obtained from the US Geological Survey. • Landuse/landcover map of Ghana (traced on Landsat TM data of 1998 and published in the same year), obtained from the Remote Sensing Application Unit (RSAU), University of Ghana, Legon. The GIS operations were carried out in three principal steps: (1) spatial data capture; (2) generation of spatial factor maps; and (c) spatial data analysis. The GIS operations were carried out using ILWIS (Integrated Land and Water Information Systems), a GIS software package developed by the International Institute for Geo-information Science and Earth Observation (ITC) in the Netherlands. ### Spatial data capture The different analog maps were scanned then georeferenced (by defining the x and y coordinates of the corner points of the maps) into a UTM coordinate system. From the scanned map, spatial data were captured by screen digitizing. From the topographic map, rivers, streams and gullies were digitised as line segments as were elevation contours. The boundaries of the district were digitised as line segments and then polygonized. The locations of centres of 61 settlements (identifiable on the topographic map) were digitised as points and the BU incidence in 1999 was recorded as spatial attribute of each settlement. From the stream sediment sample location map, the locations of the samples were digitised as points and the arsenic concentrations (in ppm) were recorded as a spatial attribute of each sample. The ASTER imagery was also georeferenced to the same coordinate system using eight reference points (tie points), which were selected in the image and which could be identified in the topographic map. Using an affine transformation, a root mean square error (RMSE) of 0.58 pixel was obtained in georeferencing the ASTER imagery. For each of the settlements with incidence of BU the percentage prevalence of BU was calculated. Prevalence expresses cases of a disease in terms of the proportion of the population afflicted at a specified time \[[@B44]\]. It is expressed here as the number of BU cases in a settlement in 1999 divided by the estimated population in 2000 multiplied by 100 to yield a percentage. ### Spatial factor maps The spatial factor maps generated from the stream sediment geochemistry data for use in the spatial analysis were: (a) map of arsenic-enriched catchment basins; (b) map of arsenic-enriched farmlands; (c) map of arsenic-enriched drainage channels. #### Arsenic-enriched catchment basins The stream sediment geochemical data for arsenic were initially analysed statistically to determine a threshold value that divides the data into background (normal) classes and anomalous (abnormally high) classes of arsenic concentrations. The data are lognormally distributed and, after removing obvious outliers in the data, a geometric mean of 8.9 ppm As and standard deviation of 2.8 ppm As were obtained. The threshold value was therefore set at 15 ppm As (i.e., approximately the mean plus two standard deviations). The spatial distribution of arsenic was then mapped through the generation of a catchment basin anomaly map in which a sample catchment basin is assigned the geochemical attribute of the corresponding sample \[[@B45],[@B46]\]. Generation of sample catchment basins involved the following steps (using ILWIS): • creation of a raster digital elevation model (DEM) through interpolation of elevation contours; • generation of raster map of drainage lines; and • calculation of sample catchment basin boundaries via an iterative calculation procedure involving the DEM and the raster map of drainage lines. The catchment basin map of arsenic concentrations was then classified into a binary map showing arsenic-normal areas (with ≤ 15 ppm As) and arsenic-enriched areas (with \> 15 ppm As) as shown in Figure [2](#F2){ref-type="fig"}. About 24% of the study area is occupied by arsenic-enriched catchment basins. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Binary map of the catchment basin.Binary map (of the catchment basin) showing arsenic-normal and arsenic-enriched areas. ::: ![](1476-072X-3-19-2) ::: #### Arsenic-enriched farmlands A supervised classification of ASTER imagery was carried out to distinguish between the major landcover/landuse classes known in the area. These landcover classes are (a) forest areas, (b) residential areas or settlements (bare of vegetation), and (c) farmlands. Using the available landuse/landcover map and topographic map as references, training pixels of known landuse/landcover classes were selected using a colour composite of ASTER bands 2, 3 and 4. These three bands gave the highest optimal index factor (OIF), which indicates the combination of three spectral bands that provide optimum information about landcover \[[@B47]\]. The box classifier \[[@B48]\] was chosen for the image classification. The classified image (Figure [3](#F3){ref-type="fig"}), which was also validated in the field, has an overall accuracy of at least 91% with reference to the landcover/landuse map. The classified image indicates that about 91% of the area is farmland. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Landcover/landuse map.Landcover/landuse map based on supervised classification of ASTER data. ::: ![](1476-072X-3-19-3) ::: To determine arsenic-enriched farmlands, a Boolean AND operation was performed by crossing the catchment basin anomaly map and the classified landcover/landuse image. About 21% of the total area of farmlands in the classified image is arsenic-enriched. #### Arsenic-enriched portions of the drainage systems A Boolean AND operation was performed by crossing the catchment basin anomaly map and the raster map of drainage lines. About 22% of the total length of drainage lines is indicated to be arsenic-enriched. ### Spatial data analysis The inhabitants of a settlement earn their livelihoods by exploiting the resources of the surrounding land. This land influences their exposure to infections and to environmental factors that dispose to infections. Proximity analysis was therefore used to determine spatial relationships between BU prevalence per settlement and (i) arsenic-enriched farmlands and (ii) arsenic-enriched portions of the drainage system. The proximity analysis was carried in two principal steps. First, maps of distances from arsenic-enriched farmlands and arsenic-enriched portions of the drainage system were generated. Second, the point map of BU prevalence per settlement was overlaid on (or crossed with) each of these maps. A buffer is a zone of specified distance around a selected map feature. A GIS creates buffer zones around selected map features such as arsenic-enriched farmlands and arsenic-enriched portions of drainage systems. Around each of these, buffers were set at intervals of 100 m up to 1000 m. Each buffer zone map was crossed with BU prevalence data of settlement to determine how many of these fall within and outside of the buffer. At each increasing interval of 100 m, a test of the significance of the difference of the mean BU prevalence within the buffer and outside of the buffer is made using the t-statistic: ![](1476-072X-3-19-i1.gif) where ![](1476-072X-3-19-i2.gif), ![](1476-072X-3-19-i3.gif) are the sample means, ![](1476-072X-3-19-i4.gif) is the pooled sample variance, n~i~and n~j~are the sample sizes from population iand j. Using t~ij~and degrees of freedom given by n~i~+ n~j~-2, a tdistribution look-up table provides the probability, Pthat the means are significantly different. Authors\' contributions ======================= AAD carried out the research and drafted the manuscript. EJMC guided parts of the research and both EJMC and MH reviewed the manuscript.	PubMed Central	2024-06-05T03:55:48.455822	2004-9-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522823/", "journal": "Int J Health Geogr. 2004 Sep 15; 3:19", "authors": [ { "first": "Alfred A", "last": "Duker" }, { "first": "Emmanuel JM", "last": "Carranza" }, { "first": "Martin", "last": "Hale" } ] }
PMC522824	Introduction ============ The control of proliferation, differentiation and survival of normal and malignant cells is under the control of a wide range of different factors. These include cell:cell interactions, immune regulatory factors, hormonal influences, and local environmental influences. However, the way in which these factors interact to regulate the dynamics of the malignant cell population are poorly understood. It is important to identify important factors and the way that they interact in order to rationalise treatment and develop new therapeutic options. However, one of the main problems is the difficulty in designing experiments to compare the effects and interactions of multiple variables. For example, a traditional experimental design to compare seven independent variables at three different concentrations each requires a large number of individual experiments (2187 experiments). The logistical and resource implications of this experimental design make these experiments very difficult to carry out. We have investigated the use of an alternative approach to experimental design, the Taguchi Methods \[[@B1]\]. Taguchi methods use orthogonal array distribution to design an experiment producing smaller, less costly experiments that have a high rate of reproducibility. A study involving 7 factors at 3 different concentrations can be conducted with only 18 individual experiments. Besides being efficient, the procedures for using Taguchi designs and methods are straightforward and easy to use. These methods have previously been used in PCR optimisation \[[@B2],[@B3]\], baculovirus expression \[[@B4]\], ball and socket prosthesis design for total hip replacement surgical procedure \[[@B5]\], ELISA optimisation \[[@B6]\], and also in the evaluation of medical diagnostic tests \[[@B7],[@B8]\]. We have therefore used a simple and reproducible assay, the MTT assay, to evaluate whether the Taguchi methods can be used to investigate the effect of G-CSF, GM-CSF, IL3 and M-CSF on daunorubicin mediated cytotoxicity in K562 cells. Taguchi Methods --------------- Taguchi methods consist of 3 phases: designing the experiment, running and analysing, and confirming and validating the assumptions. After selecting the variables to be studied, Taguchi methods depend on distributing the factors under study in an orthogonal array, which distributes the variables (factors) in a balanced manner. Examining a typical orthogonal array (Table [1](#T1){ref-type="table"}), where each factor has 2 levels or concentrations, reveals that each level has an equal number of occurrences within each column. For each column of the orthogonal distribution below, level 1 occurs four times, and level 2 occurs four times as well \[[@B1]\]. This idea of balance goes farther than meaning simply an equal number of levels within each column. The relationship between one column and another is arranged so that for each level within one column, each level within any other column occurs an equal number of times as well. With reference to Table [1](#T1){ref-type="table"}, it can be observed that factor A is assigned to column 1, and for A at level 1, factor B is at level 1 twice and at level 2 twice. The same is true for factor A at level 2. Looking at the last column, the same relationship between factors A and G is also noted. No matter which two columns are selected, the same will be true. The ramifications of this orthogonality among columns are the basis of the statistical independence of orthogonal arrays; hence the effect of each factor can be separated from the others. Therefore, an estimation of the effect of any one particular factor tends to be accurate and reproducible because the estimated effect does not include the influence of other factors. Furthermore, each factor can be assigned a significance weight to denote its importance in affecting the end result of the experiment. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Orthogonality. The relationship between one column and another is arranged so that for each level within one column, each level within any other column occurs an equal number of times as well. Factor A, at level 1 occurs 4 times and at level 2 occurs 4 times as well. This equal occurrence is true for all factors involved in any orthogonal array. ::: A B C D E F G Results --- --- --- --- --- --- --- --- --------- 1 1 1 1 1 1 1 1 Y1 2 1 1 1 2 2 2 2 Y2 3 1 2 2 1 1 2 2 Y3 4 1 2 2 2 2 1 1 Y4 5 2 1 2 1 2 1 2 Y5 6 2 1 2 2 1 2 1 Y6 7 2 2 1 1 2 2 1 Y7 8 2 2 1 2 1 1 2 Y8 ::: Each array can be identified by the form L~A~(B^C^), the subscript L, which is designated by A, represents the number of experiments that would be conducted using this design, B denotes the number of levels or concentrations within each column which denotes how many levels or concentrations could be investigated, while the letter Cidentifies the number of columns available within the orthogonal array which indicates how many factors or variables could be included in the experiment \[[@B1]\]. For example the orthogonal array L~8~(2^7^) means that 8 experimental runs are needed to investigate 7 different factors, each of which is set at 2 predetermined levels or concentrations (Table [1](#T1){ref-type="table"}). The statistical independence of these arrays enables the effect of each factor to be separated from the others, the effects to be accurate and reproducible because the estimated effect does not include the effects of other factors and the interactions between these factors to be determined. Level average analysis, as described by Taguchi \[[@B1]\] is one of the techniques used to explore the results of the Taguchi methods. The name derives from determining the average effect of each factor on the outcome of the experiment. The goal is to identify those factors that have the strongest effects and whether they exert their effect independently or through interacting with other factors. The equation below illustrates the method of calculating the average effect of the experiment where Y1 is the result of the first experiment, Y2 is the result of the second experiment\...etc, T is the overall average of the experiment, and n is the number of the experimental runs. ![](1477-7800-1-7-i1.gif) For example, in order to calculate the effect of the two concentrations of factor A, which are denoted A1 and A2, where A1 is the average effect of factor A at concentration 1, A2 is the average effect of the same factor at concentration 2. ![](1477-7800-1-7-i2.gif) The relative impact of each factor (ΔX) is simply the range, which could be calculated as the difference between the highest and lowest average response of each level. For example the impact of factor A on the experiment outcome is the difference between A1 & A2. (Known statistically as the range (Δ)). The effects of all factors are calculated in the same way, then arranged in a response table, and examined for those factors with the strongest effect (i.e. highest difference Δ), in order to separate them from the weak effects. The breaking point between the strong and weak effects is identified as a change in the pattern of the difference between the ranges around the median. Besides determining the effects of the individual factors, the same technique is used to determine the strength of the impact of interactions on the product of the experiment. The calculations are performed as the previous section. In order to determine the interactions between A and B, the average result of each 2 factors combined must be determined. This is achieved through calculating the values of 4 points: A1B1, A1B2, A2B1, and A2B2, where A1B1 is the average result generated due to the interaction between concentration 1 of both factors, A1B2 is the result of the interaction between concentration 1 of factor A and concentration 2 of factor B, A2B1 represents the interaction between factor A at concentration 2 and factor B at concentration 1, while the fourth point A2B2 is the interaction between both factors at concentration 2. These 4 points are then presented graphically to show the strength or weakness of the interaction. Whether the interaction is weak, mild, or strong depends on whether the two response lines are parallel, converging or intersecting, with intersecting lines indicate a strong interaction, and parallel lines indicate no interactions. Once the strong factors and interactions have been identified, an estimate of their combined effect is calculated and the new experiment is designed according to these assumptions. An experiment is then carried out -- referred here to as \"confirmation run\"- to validate the assumptions upon which the new experiment was based. Conducting a confirmation run and the comparison between the actual and the predicted results is necessary. If however, the confirmation results are disappointing, the planning phase must be re-evaluated and the elements that went into the experiment must be reviewed. A possible cause could be the omission of a key factor from the experiment, for example a powerful interaction was not considered. Another common cause is the setting of factor levels too close together for the experiment. In these situations, the factor is found insignificant during the analysis and is not accounted for in the validation. The confirmation run should include the best or preferred settings for mild and weak influences as well as the strong ones. However, the less influential factors are not incorporated into the prediction equation. The reasoning is that the differences in the average results may be due to experimental variation, and to incorporate their effects could result in an overestimate of the predicted results. This could lead to a disappointing confirmation run when actually the results would have validated the experiment analysis if the predicted results had not been artificially high or low. Methods ======= Cell Culture ------------ K562 cells were cultured in RPMI-1640 medium supplemented by 10% (v/v) foetal bovine serum, 50 μg/ml penicillin and 25 μg/ml streptomycin at 37°C in a humidified atmosphere of 5% CO~2~-95% air. Cells were plated in 96 well microtiter plates (200 μl) at a density of 3 × 10^4^cells/ml. Cells were co-cultured in the presence of 0.1 μg/ml daunorubicin. Cytokines --------- In all experiments K562 cells were co-incubated in the presence or absence of cytokines concentrations shown in Table [2](#T2){ref-type="table"}. All cytokines were purchased from R&D Systems, UK. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Concentrations of cytokines used. ::: 1 2 --- ------- ----------- ---------- A MCSF 100 U/ml 300 U/ml B IL3 10 ng/ml 50 ng/ml C GMCSF 10 ng/ml 50 ng/ml D GCSF 10/ ng/ml 50 ng/ml ::: MTT Assay --------- 50 μl of MTT (3--4,5-dimethylthiazol 2,5-diphenyl tetrazolium bromide) (5 mg/ml) was then added to each well and incubated at 37°C for 4 hours. The resulting deep blue crystals were dissolved in 0.04 N HCl Isopropyl alcohol, and the absorbance measured using a scanning multiwell spectrophotometer at dual wavelength 570--630 nm. All measurements were performed in triplicates. The % survival was calculated as ![](1477-7800-1-7-i3.gif) Classical Experimental Design ----------------------------- In classical experimental design the effect of each factor, each concentration and each interaction is tested independently. In order to investigate the full interactions between 4 factors each at 2 concentrations, requires 81 individual experimental conditions to be performed. This study used 49 combinations. The structure of the 49 experiments is shown in Table [3](#T3){ref-type="table"}, where runs 1--16 were designed to include all the different possible combinations of all 4 cytokines together. For example, experimental run 3 was carried out after adding 100 U/ml MCSF, 50 ng/ml of both IL-3 and GMCSF, and 10 ng/ml GCSF to the medium. Runs 17--24 included the effects of each cytokine individually; two concentrations of each cytokine were tested. For example in run 17,, only MCSF was added to the medium at concentration 1 (100 U/ml), while in run 18 the same cytokine was added at concentration 2 (300 U/ml). Runs 25--48 were planned to included the different possible interactions between each 2 cytokines, for example in run 25 both MCSF (100 U/ml) and IL-3 (10 ng/ml) were added to the medium, in run 26 MCSF (100 U/ml) and IL-3 (50 ng/ml) were added, in run 27 MCSF (300 U/ml) and IL-3 (10 ng/ml) were added, and in run 28 MCSF (300 U/ml) and IL-3(50 ng/ml) were added. Finally, experimental run 49 was carried out without adding any cytokines to the medium. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The whole set of the 49 experiments carried out. Runs 1--16 included all possible combinations of all cytokines together (see text above). In runs 17--24 individual cytokine were added to the medium, two concentrations of each cytokine was tested. For example in run 17 MCSF was added to the medium at concentration 1 (100 U/ml), while in run 18 the same cytokine was added at concentration 2 (300 U/ml). Runs 25 -- 48 included the different possible interactions between each 2 cytokines, for example in run 25 both MCSF (100 U/ml) and IL-3 (10 ng/ml) were added, in run 26 MCSF (100 U/ml) and IL-3 (50 ng/ml) were added, in run 27 MCSF (300 U/ml) and IL-3 (10 ng/ml) were added, and in run 28 MCSF (300 U/ml) and IL-3 (50 ng/ml) were added. Experimental run 49 was carried out without adding any cytokines to the medium. All experimental runs were done in triplicate and repeated three times. ::: A MCSF B IL-3 C GMCSF D GCSF ---- -------- -------- --------- -------- 1 1 1 1 1 2 1 1 2 2 3 1 2 1 2 4 1 2 2 1 5 2 1 1 2 6 2 1 2 1 7 2 2 1 1 8 2 2 2 2 9 1 1 1 2 10 1 2 1 1 11 1 1 2 1 12 1 2 2 2 13 2 1 1 1 14 2 2 2 1 15 2 1 2 2 16 2 2 1 2 17 1 0 0 0 18 2 0 0 0 19 0 1 0 0 20 0 2 0 0 21 0 0 1 0 22 0 0 2 0 23 0 0 0 1 24 0 0 0 2 25 1 1 0 0 26 1 2 0 0 27 2 1 0 0 28 2 2 0 0 29 1 0 1 0 30 1 0 2 0 31 2 0 1 0 32 2 0 2 0 33 1 0 0 1 34 1 0 0 2 35 2 0 0 1 36 2 0 0 2 37 0 1 1 0 38 0 1 2 0 39 0 2 1 0 40 0 2 2 0 41 0 1 0 1 42 0 1 0 2 43 0 2 0 1 44 0 2 0 2 45 0 0 1 1 46 0 0 1 2 47 0 0 2 1 48 0 0 2 2 49 0 0 0 0 ::: Taguchi Design L~8~(2^7^) ------------------------- In order to evaluate the performance of the Taguchi methods, eight experimental runs were carried out employing the orthogonal array L~8~(2^7^) to investigate the effect of 4 cytokines on the survival of K562 leukaemic cells. This array accommodated 4 factors MCSF, IL-3, GMCSF and GCSF. The interaction between MCSF and the other cytokines was inserted into the array. As the 3^rd^column was used to examine the interaction between MCSF and IL-3, columns 5 and 6 were used to assess the interaction between the same cytokine and GMCSF and GCSF respectively (Table [4](#T4){ref-type="table"}). All factors in this design were set at 10 and 50 ng/ml except MCSF, which was set at 100 and 300 U/ml. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Taguchi method L~8~(2^7^). This array accommodated 4 different factors (MCSF, IL-3, GMCSF, and GCSF) each at 2 different concentrations (see above). 8 experimental runs were carried out according to the combination of factors in the array, for example, in experimental run 1 the MTT assay was carried out after mixing the cells with 100 U/ml MCSF, 10 ng/ml IL-3, 10 ng/ml GMCSF, and 10 ng/ml GCSF. The interaction between MCSF and the other three factors (IL-3, GMCSF and GCSF) was studied in this array. ::: A MCSF B IL3 AxB C GMCSF AxC AxD D GCSF --- -------- ------- ----- --------- ----- ----- -------- 1 1 1 1 1 1 1 1 2 1 1 1 2 2 2 2 3 1 2 2 1 1 2 2 4 1 2 2 2 2 1 1 5 2 1 2 1 1 1 2 6 2 1 2 2 2 2 1 7 2 2 1 1 1 2 1 8 2 2 1 2 2 1 2 ::: Results and Discussion ====================== Results of the classical design ------------------------------- The results of the 49 experimental conditions are shown in Figure [1](#F1){ref-type="fig"}. Columns 1--24 represent the simultaneous combinations of the 4 cytokines. The toxicity of daunorubicin was maximally enhanced by the addition of the four cytokines to the medium i.e. MCSF at a concentration of 300 U/ml, IL-3 at a concentration of 50 ng/ml, GMCSF at a concentration of 50 ng/ml, and GCSF at a concentration of 50 ng/ml. This resulted in a highly significant reduction of malignant cell survival, from 69% (the survival rate for the control cells) to 39% (P \<0.001). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Experimental runs 1--16 show the different survival rates of K562 cells as a result of culturing the cells in medium enriched by different combinations of the 4 cytokines (GMCSF, MCSF, IL-3, and GCSF). The maximum cytotoxicity of daunorubicin was observed as a result of the addition of 300 U/ml of MCSF, and 50 ng/ml of the other 3 cytokines (experimental run 3). The maximum survival of the cells was observed when the concentration of GMCSF in this mixture was reduced to 10 ng/ml (experimental run 16). Experiments 17 -- 48 suggested that MCSF interacts with the 3 other factors to affect daunorubicin cytotoxicity. This could be seen by comparing the effect of the individual factors (runs 17 -- 24) with the effects of the addition of two factors simultaneously. For example, experimental run 30 shows the concurrent effect of both MCSF (100 U/ml) and GMCSF (50 ng/ml) that resulted in a survival that was significantly higher than that caused by any of the two factors alone (runs 17, 18, 21 & 22). Experimental run 26 also represents the combined effects of MCSF (100 U/ml) and IL-3 (50 ng/ml), which resulted in a survival that was higher than the resulting survival of any of the two factors individually. Run 36; on the other hand, represents the increase in daunorubicin cytotoxicity as a result of the simultaneous addition of MCSF (300 U/ml) and GCSF (50 ng/ml). All these experimental runs were done in triplicate and repeated 3 times, the results are expressed as mean ± SE. ::: ![](1477-7800-1-7-1) ::: The survival of K562 cells, using the 4 cytokines simultaneously was maximally enhanced by the addition of 300 U/ml MCSF, 50 ng/ml IL-3, 50 ng/ml GMCSF, and 50 ng/ml GCSF. A significant improvement in cell survival from 69% to 76% (P 0.02) was observed Taguchi analysis ---------------- The results of the 8 experiments of Taguchi\'s L~8~series (Table [5](#T5){ref-type="table"}), were analysed in order to determine the mean effect of each factor. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### the results of L~8~(2^7^). Each experimental run was done in triplicate and repeated 3 times, the mean values were calculated and the results were expressed as mean ± SE. Y1 (experimental run 1), for example, = the mean survival of the cells at 100 U/ml MCSF, 10 ng/ml IL-3, 10 ng/ml GMCSF, and 10 ng/ml GCSF. The overall average of the experiment (T) was calculated as the mean of all eight experimental runs. ::: \% survival ---- ------------- Y1 60.71 ± 5.9 Y2 67.83 ± 1.9 Y3 64.01 ± 1.1 Y4 63.51 ± 2.0 Y5 62.97 ± 4.3 Y6 72.27 ± 4.3 Y7 62.86 ± 1.1 Y8 39.84 ± 1.9 T 61.75 ::: For example the mean effect of MCSF when added to the medium at a concentration of 100 U/ml was computed as follows: ![](1477-7800-1-7-i4.gif) When MCSF was added to the medium at a concentration of 300 U/ml the mean effect was: ![](1477-7800-1-7-i5.gif) These computed values were used to construct a response table (Table [6](#T6){ref-type="table"}). This shows the average mean effect for each factor and the relative impact or range of each factor on the variability of the mean. This showed that the following concentrations were associated with lower survival of the malignant cells; 300 U/ml MCSF was associated with the survival of 59.4%, 50 ng/ml of IL-3 was associated with 57.5 % survival rate, 50 ng/m GMCSF resulted in the survival of 60.8% of the cells, and finally 50 ng/ml of GCSF was associated with 58.6% survival rate. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Response table for the orthogonal array L~8~(2^7^). The average effect of each factor level is calculated and the range of effect of each factor is calculated as the difference between the two readings. The range of MCSF effect, for example = 64.02-59.48 = 4.53, the higher the range the stronger the effect of the factor. In this experiment the interaction between MCSF and GCSF had the strongest effect on the survival of cells. ::: A = MCSF B = IL3 AxB C = GMCSF AxC AxD D = GCSF --- ---------- --------- -------- ----------- -------- -------- ---------- 1 64.02% 65.95% 57.81% 62.64% 59.21% 56.76% 64.84% 2 59.48% 57.55% 65.69% 60.86% 64.29% 66.74% 58.66% Δ 4.53 8.39 7.88 1.77 5.08 9.98 6.17 2 3 1 4 ::: In order to determine the strong effects and separate them from the weak ones, the response table was rearranged by ranking the factors in order from the largest difference to the smallest as it can be seen in Table [7](#T7){ref-type="table"}. In this study the interaction between MCSF and GCSF (AXD) has the greatest effect on the survival of K562 cells. IL-3 (B) is next with a difference between the Δ\'s of 1.594. The interaction between MCSF and IL-3 (AXB) was next followed by GCSF (D). The difference between ΔB and ΔAXB is 0.507 (8.391-7.884 = 0.507), if we continue farther to factor D the difference in effects jumps to 1.711 (7.884-6.173 = 1.711). Therefore this point would be considered as the breaking point. The factors to its left (AXD, B, and AXB) are the important factors i.e. MCSF, IL-3, and GCSF. MCSF exerts strong effects through interacting with both IL-3 and GCSF. ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Descending rearrangement of the response table according to strong and weak effects. The response table was rearranged according to the Δs, and the difference between the Δs was calculated and then scanned to determine the break point, which was identified as a change in the pattern of the difference between the Δs around the median. The strong factors would be on the left hand side of the break point, marked in this table in bold. ::: AxD B = IL3 AxB D = GCSF AxC A = MCSF C = GMCSF -------- ----------- ----------- ---------- -------- ---------- ----------- 9.985% 8.391% 7.884% 6.173% 5.088% 4.533% 1.774% 1.594 0.507 1.711 1.085 0.555 2.759 ::: In order to study each interaction incorporated into this design, an interaction matrix for each interaction was constructed as described above; hence a 2 × 2 matrix was constructed for each interaction. For example to construct an interaction matrix for MCSF and GCSF (AXD), four points were computed A1D1, A1D2, A2D1, and A2D2 (Table [8](#T8){ref-type="table"}). ::: {#T8 .table-wrap} Table 8 ::: {.caption} ###### Interaction matrix AxD. The average effect of the four points of this interaction matrix on the survival of K562 cells. The preferred setting of this interaction that would maximise the cytotoxicity of daunorubicin is A2D2 i.e. 300 U/ml of MCSF and 50 ng/ml of GCSF. This combination would result in a survival of 51.67% of the cells. ::: D1 D2 ---- -------- ------------ A1 62.11% 65.92% A2 67.56% 51.67% ::: A1D1 is the average result of the combined effect of MCSF at a concentration 100 U/ml and GCSF at a concentration of 10 ng/ml. The interaction matrix AXD was then represented graphically (Fig [2](#F2){ref-type="fig"}) which showed intersecting lines indicating a strong interaction. Both table [8](#T8){ref-type="table"} and fig [2](#F2){ref-type="fig"} were further studied to decide which combination suites the desired outcome of the experiment; hence for the smaller the better outcome, it is clear that 51.67% is the lowest survival value in this matrix (Table [8](#T8){ref-type="table"}), i.e. A2D2 (MCSF at a concentration of 300 U/ml and GCSF at a concentration of 50 ng/ml) is associated with the lowest survival rate of the malignant cells. The interaction between MCSF and the two other factors was further studied and the preferred combinations for both interactions were A2B2 i.e. 300 U/ml MCSF and 50 ng/ml IL-3, which was associated with 51.35% survival rate, and A2C2, i.e. MCSF at concentration 2 (300 U/ml) and 50 ng/ml GMCSF, which was associated with 56.05% survival. The preferred settings suggested by this analysis to optimise the cytotoxicity of daunorubicin was combining the following factors A2B2C2D2 i.e. MCSF at a concentration of 300 U/ml, IL-3 at a concentration of 50 ng/ml GMCSF at a concentration of 50 ng/ml, and GCSF at a concentration of 50 ng/ml. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Graphical presentation of interaction between AxD (MCSF and GCSF). Intersecting lines of this graph indicate strong interaction. A2D2 is the preferred point on the graph i.e. the combination of these two factors to produce maximum daunorubicin cytotoxicity. ::: ![](1477-7800-1-7-2) ::: An estimate of the predicted response (μ) based on the selected levels was then computed. The calculations were based on the overall average value (T) and the effect that each of the recommended levels of the strong factors and interactions has on the overall average. μ = T+(A2D2 - T)+(B2 - T)+(A2C2 - T)+(A2B2 - T)-(A2 - T)-(A2 - T)-(A2 - T)-(B2 -T)-(C2-T). The reason for subtracting the individual effects of factors A, B, and C from the effects of A2B2 is that A2B2 is comprised of the effects of factor A, factor B and the interaction itself. Unless the effects of the two factors are subtracted these strong effects would be included twice and resulting in an overestimation of the predicted result. The predicted survival derived from the above prediction equation was 43%. A confirmation run that produces a %survival close to 43%would validate the assumptions of this Taguchi method. The actual confirmation run, in fact, resulted in 39.84% survival rate indicating the success of the Taguchi experiment. Further calculations were performed to determine whether the outcome of other combinations could be predicted from the Taguchi experiment and confirmed by analysis. The results are shown in Table [9](#T9){ref-type="table"} and show a close approximation in each case. ::: {#T9 .table-wrap} Table 9 ::: {.caption} ###### Further comparison of the predicted values from the Taguchi Methods, and the result produced by experimental analysis. ::: Prediction Analysis --------------------- ------------ ---------- Experimental run 9 66.10% 70.35% Experimental run 10 71.93% 74.38% Experimental run 11 66.14% 63.36% Experimental run 12 67.65% 65.36% Experimental run 13 48.90% 45.69% Experimental run 14 51.46% 55.22% Experimental run 15 63.65% 74.88% Experimental run 16 72.19% 76.74% ::: Conclusion ========== The aim of this study was to evaluate the ability of the Taguchi methods to investigate the effects of several factors simultaneously on the death and/or survival of the malignant cells, and to compare this strategy against a traditional full experimental design. A major finding of the study was that the Taguchi methods predicted the combination of factors that results in the lowest survival of the malignant cells. This agreed with the conclusions of the full experimental design but required only eight individual experiments to pinpoint this combination. However, it must be stressed that the Taguchi methods are not intended to be a replacement for traditional experimental design, but if used as a complimentary strategy can make analysis of complex interactions feasible and practical. In this study, for example, eight individual experiments produced a testable combination which required 49 individual experiments to produce in the traditional experimental design. In more complex systems, only Taguchi methods become feasible. For example, to study 13 factors at 3 different combinations would require 1,594,323 individual experiments at a cost for re agents alone of over 27 million pounds. If Taguchi methods are used, this can be reduced to just 27 individual experiments at a cost of under 500 pounds. We have described a novel experimental approach to studying the interactions of several factors on the cytotoxicity of malignant cells. We show that the method is effective in the determination of the optimum conditions, even in the presence of multiple interactions. We anticipate that this experimental strategy will have many applications in the investigation of complex interactions. For example we have used this strategy to model the complex testicular microenvironment and the ability to support the survival of acute lymphoblastic leukaemia cells (manuscript in preparation). These sort of interactions are common in the survival of malignant cells in vivo, and we propose that the Taguchi methods may be a useful strategy to understand these interactions in vitro, and to help devise and implement new therapeutic strategies. Competing interests =================== None declared. Authors\' contributions ======================= HM designed the Taguchi assay, carried out cell survival assays, and drafted the manuscript. KLY participated in the design and coordination, and produced the final manuscript. APJ conceived the study and participated in its design and coordination. All authors read and approved the final manuscript.	PubMed Central	2024-06-05T03:55:48.457900	2004-9-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522824/", "journal": "Int Semin Surg Oncol. 2004 Sep 20; 1:7", "authors": [ { "first": "Hisham", "last": "Morsi" }, { "first": "Kwee L", "last": "Yong" }, { "first": "Andrew P", "last": "Jewell" } ] }
PMC522825	Background ========== Poor diet is thought to be a risk factor for many diseases \[[@B1],[@B2]\]. One way of evaluating this risk is to carry out studies using dietary assessment techniques. Food frequency questionnaires (FFQ) have been the primary method of food self-reporting in nutritional epidemiology for the past 20 years, but it is now suggested that the ability to study associations between diet and chronic diseases may be better served by using a food diary \[[@B3]\]. The most accurate methods for dietary assessment are direct observation in the home, or a food history, which involves a 1--2 hour interview by a specially trained nutritionist. These methods are costly and the food diary is often used when they are not possible \[[@B4]\]. Age-related macular degeneration (AMD) is the leading cause of blind registration and visual disability in patients over the age of 60 years in the developed World \[[@B5]\]. The condition affects more than 1.75 million people in the United States, and it is expected that the demographic right-shift will lead to an increase in this number to almost 3 million by 2020 \[[@B6]\]. In accordance with the International Classification and Grading System for Age-Related Maculopathy (ARM), and Age-Related Macular Degeneration (AMD), these abbreviations will be used throughout \[[@B7]\]. The term age-related macular disease (ARMD) will be used to encompass ARM and AMD. ARM is the early stage of ARMD and is most often clinically apparent over the age of 50 years. The main symptom is increasing difficulty with fine detail discrimination. AMD is the later stage of ARMD and is categorised further in to \'dry AMD\' (also known as geographic atrophy, GA), and \'wet AMD\' (also known as \'neovascular\', \'exudative\', or \'disciform\' AMD) \[[@B7]\]. GA is the most common form, and is estimated to be present in 15% of eyes by 80 years of age \[[@B8]-[@B11]\]. Progression is slow and legal blindness has been estimated to occur between 5 and 10 years \[[@B12]\]. Exudative AMD is less common, occurring in 5.2% of the population over 75 years \[[@B13]\], but accounts for a 90% blind registrations\[[@B14]\]. Patients experience rapid, significant loss of central vision as a result of growth of new blood vessels beneath the retina. The prevalence of GA and exudative AMD in the US population over 40 years of age has been estimated at 1.47% \[95% confidence interval (CI), 1.38% -- 1.55%\] \[[@B6]\]. The likelihood of visual deterioration in those with exudative AMD may be reduced with laser treatment \[[@B15]-[@B18]\], although success is limited. The paucity of treatment options has prompted interest in the identification of risk factors, as well as the development of prevention strategies. The three main risk factors are increasing age \[[@B19]-[@B26]\], smoking \[[@B22],[@B27]-[@B29]\], and genetic predisposition \[[@B30]-[@B34]\], although other proposed factors include gender \[[@B35],[@B36]\], race \[[@B37]-[@B39]\], socioeconomic factors \[[@B21],[@B40]\], cardiovascular disease \[[@B21],[@B31],[@B41],[@B42]\], and poor nutrition \[[@B43]-[@B45]\]. It is thought that people with low systemic antioxidant levels may be more prone to oxidative damage of the retina and therefore, AMD \[[@B46]\]. Oxidation refers to removal of electrons and is mediated by reactive oxygen intermediates (ROI), which include free radicals, hydrogen peroxide, and singlet oxygen. Free radicals are molecules that contain one or more unpaired electrons in their outer orbits \[[@B47]\], and they extract electrons from other molecules in order to achieve stability. These molecules are rendered unstable by the interaction and a cytotoxic chain reaction results. This damage is thought to contribute to the pathogenesis of many diseases \[[@B1],[@B2]\]. The hypothesised role of oxidation in the development of AMD has prompted research into the use of nutritional supplementation \[[@B48]\]. The Age-Related Eye Disease Study (AREDS) found a significant odds reduction for the development of advanced AMD with antioxidant plus zinc supplementation \[[@B49]\], and the Lutein Antioxidant Supplementation Trial (LAST) reported that visual function in AMD patients is improved with supplementation of lutein and lutein combined with other nutrients \[[@B50]\]. Lutein and its isomer zeaxanthin are carotenoids, and are synthesised in plants, algae, and bacteria. In mammalian systems they can only be obtained from the diet \[[@B51]\]. Their selective absorption by the retina, in particular the macula, is suggestive of a protective function, and has prompted use of the term macular pigment (MP) to describe them within the retina. Lutein and zeaxanthin are believed to protect the retina in two ways. Firstly, they act as blue-light filters. Action spectrum for blue-light induced damage shows a maximum between 400 nm and 450 nm, and this is consistent with the absorption spectrum of macular pigment \[[@B52]\]. Secondly, they are able to quench free radicals. Energy transfer to them quenches singlet oxygen, and they are also believed to react with peroxyl radicals that are involved with lipid peroxidation \[[@B53]\]. The primary aim of this study was to evaluate the dietary food intake of three subject groups: participants under the age of 50 years without ARMD (U50), participants over the age of 50 years without ARMD (O50), and participants with ARMD. The secondary aim was to obtain information on nutritional supplement usage. Methods ======= Study design ------------ Prospective cross-sectional in a clinical practice setting. Participants ------------ Seventy-four participants gave informed consent to take part in this study, which was approved by the Institutional Human Ethics Committee. Recruitment methods included sending information to Birmingham optometrists, ophthalmologists, and a specialist centre for rehabilitation of people with sight loss, an editorial in a local newspaper, recruitment e-mails sent to the Royal National Institute for the Blind (RNIB) and all staff and students at Aston University and Aston Science Park, Birmingham, UK. For analysis the participants were divided into three groups: U50; 20 participants aged \< 50 years, from 21 to 40 (mean ± SD, 37.7 ± 10.1 years), O50; 27 participants aged \> 50 years, from 52 to 77 (62.7 ± 6.8 years), and ARMD; 27 participants aged \> 50 years with age-related macular disease, from 55 to 79 (66.0 ± 5.8 years). All participants were part of a larger study investigating the effects of nutritional supplementation on visual function in normal and diseased eyes \[[@B54]\]. Chi squared analysis for gender yielded no significant difference between groups U50 and O50 \[χ^2^(1) = 0.104 p = 0.305\] and groups O50 and ARMD \[χ^2^(1) = 3.814 p = 0.051\]. The difference in age is not significant between groups O50 and ARMD (t = -1.842, p = 0.071). Exclusion criteria were the presence of an ocular condition other than ARMD and the presence of medical conditions indicating a diet in which particular foods or food groups were excluded (e.g. coeliac disease). Participants were issued with a three-day food diary with verbal and written instructions explaining that they should add to their diary every time they eat or drink, describing the food as accurately as possible and giving estimates of amounts. They were also asked to provide details of any daily nutritional supplements. The diary consisted of two week days and one weekend day. The diaries were analysed using FoodBase 2000 software (The Institute of Brain Chemistry and Human Nutrition, 166--220 Holloway Road, London N7 8DB, UK), which is a computerised nutrition database containing data on approximately 3750 foods. It can be used for recipe analysis, meal analysis, and daily or weekly analysis of menus or food intakes. Data were input by one investigator and statistically analysed using Microsoft Excel for Microsoft Windows XP software, employing unpaired t-tests. Results ======= Dietary analysis ---------------- The values for energy and nutrient intakes for all participants are shown in table [1](#T1){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Daily mean and SD values for energy and nutrient intake. ::: Group U50 (mean ± SD) n = 20 Group O50 (mean ± SD) n = 27 Group ARMD (mean ± SD) n = 27 -------------------------- ------------------------------ ------------------------------ ------------------------------- Energy (kcals) 1672.30 ± 425.58 1599.78 ± 331.50 1823.37 ± 546.18 Protein (g) 71.91 ± 27.62 68.14 ± 17.08 85.25 ± 18.28 Fat (g) 65.82 ± 27.77 58.44 ± 44.27 66.73 ± 21.76 Carbohydrate (g) 203.13 ± 39.46 204.43 ± 44.27 222.05 ± 82.66 Alcohol (g) 4.31 ± 4.38 4.83 ± 8.70 6.86 ± 9.54 Fibre (g) 13.04 ± 4.60 16.21 ± 5.44 17.31 ± 5.47 Cholesterol (mg) 192.19 ± 122.99 203.04 ± 82.63 242.60 ± 72.46 Zinc (mg) 8.43 ± 2.98 8.50 ± 2.58 10.07 ± 2.45 Copper (mg) 1.10 ± 0.52 1.21 ± 0.51 1.43 ± 0.54 Selenium (μg) 62.82 ± 75.84 105.01 ± 126.05 72.01 ± 62.76 Riboflavin (mg) 9.90 ± 35.82 1.57 ± 0.43 1.77 ± 0.44 Vitamin C (mg) 79.01 ± 40.67 140.59 ± 75.23 114.91 ± 60.01 Vitamin E (mg) 6.33 ± 3.49 6.71 ± 3.13 7.87 ± 3.52 Vitamin D (μg) 3.33 ± 3.06 3.05 ± 1.91 3.76 ± 2.54 Retinol equivalents (μg) 681.25 ± 499.55 679.44 ± 237.65 825.22 ± 440.54 \% energy from fat 34.20 ± 7.34 32.56 ± 4.69 32.89 ± 5.60 ::: ### Comparing group U50 with group O50 Group O50 consumed significantly more vitamin C (t = 3.049, p = 0.005) and significantly more fibre (t = 2.107, p = 0.041) than group U50. The mean intakes for men and women in each group are shown in table [2](#T2){ref-type="table"}. The data has been broken down into male/female subgroups because reference nutrient intake (RNI) data can differ with gender. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Mean vitamin C and fibre daily intake for groups U50 and O50 ::: Mean vitamin C intake (mg) Mean fibre intake (g) ------------------------------- ---------------------------- ----------------------- Women under 50 years (n = 12) 77.48 ± 42.79 12.19 ± 4.06 Women over 50 years (n = 22) 147.13 ± 75.12 16.03 ± 4.92 Men under 50 years (n = 12) 78.68 ± 36.93 14.18 ± 4.78 Men over 50 years (n = 7) 125.87 ± 78.20 15.96 ± 6.48 ::: By tradition, investigators consider a study to be adequately powered if it has an 80% chance of detecting a significant difference when one exists. The number of study participants needed to detect a clinically important difference with acceptable power, can be calculated using the required power, the expected variability of the outcomes, and the chosen probability of masking a type 1 error \[[@B55]\]. Power analysis shows that 20 subjects is not sufficient to have an 80% chance of detecting a difference of 25% or more of the mean value at the 5% level of significance using the unpaired t-test for alcohol, copper, cholesterol, selenium, vitamin E, vitamin D, and retinol equivalents. In other words, for these dietary consitiuents we cannot state whether we found no difference between groups because there actually was no difference, or because the study did not have enough power to detect a difference. The study was however, powered to assess the difference in means for energy, protein, fat, carbohydrate, zinc, riboflavin, and percentage energy from fat, and no significant differences were found. ### Comparing group O50 with group ARMD Group ARMD consumed significantly more protein (t = 3.487, p = 0.001) and zinc (t = 2.252, p = 0.029) than group O50 (see table [3](#T3){ref-type="table"}). Power analysis shows that 27 subjects is not sufficient to have 80% chance of detecting a difference in means of 25% at the 5% level of significance using the unpaired t-test for alcohol, copper, cholesterol, selenium, vitamin E, vitamin D, and retinol equivalents. The study was powered to assess the difference in means for energy, fat, carbohydrate, fibre, riboflavin, and percentage energy from fat, and found no significant difference. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Mean zinc and protein daily intake for groups O50 and AMD ::: Mean zinc intake (mg) Mean protein intake (g) ------------------------------ ----------------------- ------------------------- Women over 50 years (n = 22) 8.30 ± 2.50 66.35 ± 16.12 Men over 50 years (n = 7) 9.25 ± 2.50 71.61 ± 17.46 Women with ARMD (n = 13) 9.60 ± 2.26 78.05 ± 15.98 Men with ARMD (n = 14) 10.51 ± 2.54 91.93 ± 17.73 ::: Baseline nutritional supplement intake -------------------------------------- The results indicate that group O50 (mean ± SD; 1.44 ± 1.79) consumes significantly more types of nutritional supplement than group U50 (0.55 ± 1.11) (t = 2.220, p = 0.032). No difference found between groups O50 and ARMD, and the study was powered to have an 80% chance of detecting a difference in means of 1 at the 10% level of significance. The percentage of supplements taken for each group is shown in figure [1](#F1){ref-type="fig"}. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Daily nutritional supplement use by group. ::: ![](1475-2891-3-16-1) ::: Discussion ========== The aim of this study was to evaluate the dietary intakes of three subject groups; U50, O50 and ARMD, as well as to obtain information on nutritional supplement usage. Participants under the age of 50 years consumed significantly less dietary vitamin C than those aged over 50 years. Supplementation data shows that 7.4 % of the O50 group take uncombined vitamin C compared with 0% of the U50 group. However, a higher percentage of the U50 group take multivitamins (33.3%) compared with the O50 group (22.2%). Vitamin C is water-soluble, is involved with several biological processes. As a reducing agent it is thought to be active in protection against heart disease. It protects LDL (low density lipoprotein) cholesterol from oxidative damage and reduces platelet aggregation \[[@B56]\]. By enhancing nitric oxide activity, vitamin C is potentially important in lowering blood pressure \[[@B57]\]. High dose supplementation with an antioxidant and zinc formulation, including vitamin C was associated with a 25% reduced risk of progression of AMD in those participants already suffering with the condition \[[@B49]\]. Some studies, however, have found no evidence for a beneficial role of vitamin C supplementation in ocular disease. There was no relationship between cataract prevalence and vitamin C intake in two studies \[[@B58],[@B59]\], and no relationship between cataract extraction and vitamin C intake in a third \[[@B60]\]. Although the antioxidant properties of vitamin C are well known, there is no clinical evidence suggesting that supplementation with vitamin C can reduce the risk of ARMD, or other ocular conditions such as cataract and glaucoma. The RNI for men and women over the age of 18 years is 40 mg. The mean intakes of men and women in the U50 and O50 groups are all above this value (table [2](#T2){ref-type="table"}). The higher intake values of the O50 group may be explained by their increased awareness of the benefits of a balanced diet, consumption by this group of more traditional, home prepared foods, and lower consumption of convenience foods. An increased consumption of convenience foods in the U50 group may also explain why they consume significantly less fibre than the O50 group (t = 2.107, p = 0.041). Interestingly, all three groups had a mean intake value of less than 18 g, the RNI for fibre in men and women. The ARMD group consumed significantly more dietary zinc than age- and gender- matched controls. Zinc has been investigated with regard to its potential preventative role in ARMD. The AREDS group found a suggestive reduction in the risk of progression of AMD in participants supplementing with 80 mg zinc daily. Previous randomized controlled trials (RCTs) using 200 mg zinc daily found conflicting results \[[@B61],[@B62]\], and the positive result reported by Newsome et al(1988) should be treated with caution \[[@B48]\]. The higher intake by ARMD participants may be explained by their awareness of research into zinc supplementation and the condition. The RNI for women over 18 years is 7.0 mg and for men over 18 years is 9.5 mg. Our results show that the mean intakes were above RNI values for all four groups. Supplementation data shows that 11.1 % of the ARMD participants supplemented with zinc, compared with 3.7 % of the O50 group, and 0 % of U50 participants. The Food Standards Agency released a report on the safety of vitamins and minerals in May 2003 and suggested a safe upper limit of 42 mg for total daily zinc intake. Zinc supplementation over 150 mg/day has been associated with gastrointestinal side effects such as cramping and nausea, as well as lethargy and blood in the urine \[[@B63]\]. Our results show that the ARMD participants are most at risk of exceeding the safe upper limit as they have the highest dietary and supplemental zinc intake. The ARMD group consumed significantly more protein than O50 participants. We are not aware of any investigation into a link between protein and risk of ARMD, and table 4 shows that the mean intakes are above the RNI for both men (55.5 g) and women (46.5 g). Previous studies have found a relationship between higher dietary fat intake and risk of ARM (RR 1.6) \[[@B64]\], and high serum cholesterol and increased risk of exudative AMD compared with low serum cholesterol levels \[relative risk (RR) 4.1\] \[[@B40],[@B65]\]. However, the NHANES I found that subjects with high cholesterol intake were less likely to develop AMD than those with lower intake \[odds ratio (OR) 5.1\] \[[@B21]\]. Our results show that ARMD participants consumed more fat and cholesterol than the O50 group, although these differences were not statistically significant. The study was underpowered for cholesterol. Research into the role of alcohol consumption in the development of AMD has produced conflicting results. Several studies have found no relation \[[@B40],[@B66]-[@B70]\], but consumption of beer has been related to an increased risk of retinal pigmentation (OR 1.13) and exudative AMD (OR 1.41) \[[@B71]\]. Both men (RR 2.16) and women (RR 2.20) in the highest category of wine intake (2 or more glasses per day) have been shown to be at increased risk of AMD \[[@B67]\]. This association was strongest with white wine, and interestingly the NHANES I determined that red wine is associated with a lower risk of AMD \[[@B21]\]. This may be related to the antioxidant properties of the phenolic compounds within red wine \[[@B72]\]. Our data shows that the ARMD group consumed more alcohol than both the U50 and O50 groups, although these differences were not statistically significant and the study was underpowered for alcohol. The non-significant differences found between groups for alcohol, copper, cholesterol, selenium, vitamin E, vitamin D, and retinol equivalents may have occurred because there truly was no difference, or because the study had insufficient power to detect a difference. Because of the variability of the data, subject numbers required per group for 80 % power at the 5 % significance level are 467 for alcohol, 50 for copper, 44 for cholesterol, 341 for selenium, 59 for vitamin E, 113 for vitamin D, and 71 for retinol equivalents. Multivitamins were the most commonly taken supplement by the U50 group (30.0 %), compared with cod liver oil for the both the ARMD group (33.3 %) and O50 participants (22.2 %). Seventy-five percent of the specific ocular health related supplements were taken by the ARMD group, 25 % by the O50 group, and 0 % by the U50 group. Conclusion ========== We did not detect a deficiency of any specific nutrient in the diets of those with ARMD compared with age- and gender-matched controls. A higher percentage of ARMD participants consume specific ocular health nutritional supplements (33.3 %) compared with age- and gender-matched controls (11.1 %) and U50 participants (0 %). The U50 group consumed a higher percentage of multivitamins, but significantly less vitamin C and fibre than the O50 group. This suggests that the younger age-group might use supplementation to ensure adequate consumption of vitamins and minerals. The ARMD group consumed more dietary zinc, more supplemental zinc, and the highest percentage of ocular health related supplements. This may suggest that information regarding the results of studies investigating the role of nutritional supplementation in reducing the risk of onset or progression of AMD is reaching patients. These results however, may be confounded by the fact that the ARMD participants used in this study were enrolled in an RCT investigating the use of nutritional supplementation in ARMD. Participants in research projects may be more aware of scientific developments and more likely to investigate their condition and potential therapies. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= HB participated in the design of the study, carried out data collection, input, and analysis, and drafted and developed the manuscript. FE participated in the design of the study and development of the manuscript. Both authors read and approved the final manuscript.	PubMed Central	2024-06-05T03:55:48.461283	2004-9-28	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522825/", "journal": "Nutr J. 2004 Sep 28; 3:16", "authors": [ { "first": "Hannah", "last": "Bartlett" }, { "first": "Frank", "last": "Eperjesi" } ] }
PMC522826	Background ========== There is increasing evidence that nutrients may be important in the development of chronic diseases such as coronary heart disease (CHD) and type 2 diabetes. In the late 60s, the Mediterranean diet became a topic of interest primarily because of results of the Seven Countries Study, which demonstrated that the 15-y mortality rate from CHD in Southern Europe, was two to threefold lower than in Northern Europe or United States \[[@B1]\]. More recently, results from The Lyon Diet Heart Study showed that a Mediterranean alpha-linolenic acid-rich diet prevented the recurrence of cardiovascular events more than did the usual prudent Western diet in men \[[@B2]-[@B4]\]. Reliable instruments for diet measurements are necessary to identify which components of the Mediterranean diet are the best candidates to explain, in part, such a protective effect. Accurate assessment of dietary intakes, when based on self-report in free-living populations poses significant scientific challenges. All standard dietary assessment methods including food records, dietary recalls and list-type methods such as food frequency questionnaires (FFQ), are subjected to considerable error and bias, and none of these can be considered as a \'gold standard\' measure \[[@B5]\]. FFQ has become a common way to estimate usual food intake because it usually requires less than thirty minutes to complete \[[@B6]\]. It also imposes less burden on subjects than most of the other dietary assessment methods. However, disadvantages of the FFQ have been identified. In fact, it may be difficult cognitive task for respondent to recall frequencies of intakes over a given period of time. Also, the precision in quantifying intakes is not possible with a FFQ. Dietary habits vary not only from country to country but also from region to region. Specific FFQs must be validated to assess nutritional habits conducted in geographically and/or culturally distinct regions \[[@B6]\]. It is also important in nutritional intervention to consider the sensitivity of the method over the duration of a study, especially in study that is testing the effects of dietary changes \[[@B7]\]. In Québec, no validated FFQ was adapted for the needs of our nutritional intervention design. In fact, in the context of our nutritional intervention we wanted to use a FFQ to evaluate a Mediterranean food pattern in a North-American context that would contain foods available in Québec and also foods characteristic of the Mediterranean diet. In order to improve the precision in quantifying reported intakes we decided to use and interviewer-administered FFQ to facilitate the determination of portion size using food models. Thus, we decided to design an interviewer-administered FFQ to assess the dietary changes among a French-Canadian population in a nutritional intervention promoting the Mediterranean food pattern. The first purpose of the present study was therefore to test the validity of this interviewer-administered FFQ. In order to reach this objective, nutrient intakes derived from the FFQ were compared to intakes obtained by the 3-day food record. Comparisons were performed for baseline as well as for post-intervention values. This allowed us to determine whether our FFQ would permit to identify similar changes in nutrient intake in response to the intervention as the ones measured by the 3-day food record. As a second objective, we also wanted to estimate the short-term reproducibility of this FFQ in a control group who did not receive the nutritional intervention. Subjects and methods ==================== Subjects -------- This paper reports results of two studies: 1) a study on validity of the FFQ tested against a 3-day food record during a nutritional intervention program promoting the Mediterranean food pattern and 2) a study on reproducibility of the FFQ in a control population. For the validation study, women from the Québec City metropolitan area were recruited through the Laval University newspaper during the summer of 2001. Women included in the study were aged between 30 to 65 years \[[@B8]\]. To be eligible, women had to be free from metabolic disorders requiring treatment, to have stable body weight for at least 3 months prior to the start of the study and to be in charge of food purchases and meal preparation most of the time. One hundred and twenty six women were invited to a screening visit for an evaluation of their food habits. Among this initial group of women, 94 were found to be eligible according to the above criteria and 77 women agreed to take part to the study. Three women left the study during the 12-week intervention for personal reasons. One participant did not complete the FFQ at week 12 and 2 other participants did not complete all three food records. Therefore, 71 women were included for the FFQ validation analysis. For the reproducibility study, 20 men and 19 women from the Québec City metropolitan area were also recruited through the Laval University newspaper during the summer of 2002. Men and women included in the study were aged between 25 to 70 years. Three men and 4 women did not complete the second FFQ for personal reasons. Therefore, 17 men and 15 women were included for the reproducibility analyses. The study was approved by the Ethics Committees of Laval University. Food frequency questionnaire ---------------------------- The interviewer-administered FFQ developed inquired on the food habits during the last month and is based on typical food items, which are available in Québec. It contains 91 items among which 27 had between 1 and 3 subquestions. The FFQ was structured to reflect Quebecers\' food habits and food items were listed in food groups (vegetables; fruits; legumes, nuts and seeds; cereals and grain products; milk and dairy products; meat/processed meat; poultry; fish; eggs; sweets; oils and fats; fast foods and drinks). Because of the nature of our nutritional intervention, our FFQ was designed to make sure to document with enough details consumption of typical items of the Mediterranean diet such as type of oils, whole grain products and legumes. The 30-minutes FFQ was administered face-to-face by one of the 3 registered dietitians involved in the study. During the interview, the dietitians used food models for a better estimation of the real portion consumed by the subject. Participants were questioned about frequency of intake for different foods during the last month and were asked to report the frequency of these intakes in terms of day, week or month. The subquestions allowed a better definition of the food items consumed. For example, following the question \"How often do you eat yogurt?\" subjects were asked about the fat percentage and brand of the yogurt consumed. An open question at the end of the FFQ allowed individuals to report any other frequency eaten foods not listed in the FFQ and provide details about usual recipes used in order to quantify better intakes of individual food items. Cut-off value to evaluate energy intake --------------------------------------- Estimates of basal metabolic rate (BMR) were calculated from the Harris-Benedict formulas based on height, weight, age and sex \[[@B9]\]. Energy intake reported from FFQs and from 3-day food records were compared with estimates of BMR to calculate the number of participants who may have underreported their energy intake \[[@B10]\]. It is suggested that a ratio between energy intake and estimate BMR of less than 1.35 might not represent long term habitual intake in a non dieting population \[[@B10]\]. Study 1: Validation ------------------- ### Intervention and FFQ The methodology of the nutritional intervention has been described previously \[[@B8]\]. Briefly, the study was conducted in 2 phases. Each phase was conducted using a similar 12-week intervention design. The FFQ was administered at screening (t = 0) and then at weeks 6 and 12. The intervention included 2 group sessions. Individual sessions took place during the 1^st^, the 6^th^and 12^th^weeks of the intervention in order to evaluate the changes and to select further objectives for increasing the adherence to the Mediterranean food pattern. Three registered dietitians were trained to provide a standardized intervention. ### 3-day food record Each participant completed a 3-day food record, 2-week days and 1-weekend day, at week 0, 6 and 12. At screening of the nutritional intervention, a dietitian provided 15 minutes of instructions to each participant on how to complete the food records. Written copies of record examples were provided to each subject. Also, participants were encouraged to consume usual amounts of typical foods and drinks for the completion of the food record. Participants were not required to weight foods but were asked to measure the volume of foods consumed with household measurements (cups, tablespoons) or to indicate the weight of commercial products when it was possible to assess portion sizes. The food record included a section for recording information recipes. After completing the food record, participants met with the dietitian to review all the information for record accuracy and completeness and portion size of individual items on the food record were reviewed when needed by using food models. ### Anthropometry At weeks 0, 6 and 12, body weight and height were measured according to the procedures recommended at the Airlie Conference on the Standardization of anthropometric measurements \[[@B11]\] and body mass index (BMI) was calculated. Study 2: Reproducibility ------------------------ ### Study design The 32 participants of the reproducibility study were distributed into two groups. In the first group, dietitian \#1 administered the FFQ and 4 weeks later dietitian \#2 administered the FFQ for the second time. Inversely, in the second group of participants, dietitian \#2 administered the first FFQ and 4 weeks later dietitian \#1 administered the FFQ for the second time. An interval of one month was chosen to reduce any training effect and memory influence of the method. Both dietitians were taught to use the FFQ similarly, using the same examples of portion size, and asking similar questions. ### Anthropometry At the first visit, body weight and height were measured according to standard procedures \[[@B11]\] and BMI was calculated. Analysis -------- ### Nutritional analysis Evaluation of nutrient intakes derived from the FFQs and food records was performed using the Nutrition Data System for Research (NDS-R) software version 4.03, developed by the Nutrition Coordination Center, University of Minnesota, Minneapolis, MN, Food and Nutrient Database 31, released in November 2000 \[[@B12]\]. This database includes more than 16 000 food items for which the complete nutritional value of 112 nutrients is included. For the purpose of our study, intakes of selected nutrients susceptible to affect the CHD risk profile were analysed: energy, proteins, carbohydrates (CHO), lipids, saturated fatty acids (SFA), monounsaturated fatty acids (MUFA), polyunsaturated fatty acids (PUFA), trans fatty acids, eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), total as well as insoluble and soluble dietary fibers, vitamin C, folate, iron and calcium. Intakes of vitamin and mineral supplements were not included in the present analysis, which focused on dietary nutrients only. ### Statistical analysis In the validation study, means and standard deviations for nutrient intakes were calculated from the FFQs and from the food records. Student t-tests were performed to determine the differences between nutrient intakes assessed by the FFQ and by the 3-day dietary record. Since many variables were not normally distributed, Spearman correlations were used to put into relationship nutrient intakes from FFQ with those from the food record. Student t-tests were also performed to determine the differences between changes in nutrient intakes in response to the nutritional intervention assessed by the FFQ and by the 3-day dietary record. Agreement analyses were performed to verify the concordance of different nutritional variables among quartiles of the distribution between FFQ and the 3-day food record. In the reproducibility study, Student t test was performed to assess the differences between both FFQs. Spearman correlations were used to put into relationship the nutrients reported in the first and the second FFQ. Agreement analyses were assessed to verify the degree of concordance in classifying subjects among quartiles of the distribution between both FFQs. For both studies, the dietary variables were log transformed when necessary to achieve a normal distribution, and the formula log(x + 1) was used for alcohol because some subjects had a value of 0 g for alcohol intake. In addition, to make the comparisons based on absolute nutrient intakes, correlations were also made using energy-adjusted variables. Adjustment for total energy intake was achieved by using the residual method proposed by Willett and Stampfer \[[@B13]\]. Residuals are computed from regression models, with total energy intake as the independent variable and nutrient intakes as the dependent variable. Values were considered as being very well correlated for correlations ranging between 0.7 to 0.9, well correlated for correlations ranging between 0.5 to 0.7 and moderately well correlated for correlations between 0.3 to 0.5, as suggested by Rimm et al \[[@B14]\]. Because age and BMI may have influenced the manner in which subjects answered the FFQ or completed the 3-day food record, partial correlations for age and BMI were also computed. Also, Student t-tests were performed to verify variation in nutrients intakes obtained from the two FFQ administered by the two interviewers. All analyses were performed with the SAS statistical package version 8.02 (SAS Institute, Cary, N.C., USA). Results ======= Study 1: Validation ------------------- Women had a mean BMI of 25.8 ± 3.9 kg/m^2^and a mean age of 46.8 ± 7.9 y. Average daily nutrient intakes derived from FFQ and the 3-day food records are shown in Table [1](#T1){ref-type="table"}. Total energy intake measured by the FFQ was not different from the intake assessed by the 3-day food record (difference below 5%) at week 0 of the dietary intervention (Table [1](#T1){ref-type="table"}). For FFQ and the 3-day food record respectively, 34% (n = 24) and 38% (n = 27) of subjects had at baseline a ratio between energy intake and estimated basal metabolic rate at or below 1.35. Also, the intakes of proteins, CHO, SFA, PUFA, trans fatty acids, cholesterol, alcohol, vitamin C, folate, calcium and iron measured by the FFQ and by the 3-day food record at week 0 did not differ significantly. Measurement of total dietary fibers and soluble fibers differed significantly between the FFQ and the food-record but differences were below 10%. Mean nutrient intakes measured with FFQ were within 10% of values obtained with the 3-day food record for 14 of the 19 nutrients measured at week 0. For total lipids and MUFA the differences were significant but below 15%. Adjusting for energy intake did not alter these observations. Similar observations were noted at week 12 except for intakes of lipids, total dietary fibers and soluble fibers that were not anymore significantly different between the two methods and for PUFA intake that was estimated as being significantly higher with the FFQ (data not shown). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Mean values of daily intakes of nutrients and Spearman correlation coefficients between values derived from the 3-day food record and the FFQ at week 0 of the nutritional intervention (n = 71). ::: Dietary record FFQ Difference (%)^a^ Unadjusted Energy-adjusted ------------------------- -------------------- ----------------- ----------------------- ---------------- --------------------- Energy (kcal)^b^ 2055 ± 521 2143 ± 568 4.3 0.29\\ Protein (g) 81.3 ± 16.4 82.6 ± 23.3 1.6 0.27\* 0.36\\ CHO (g) 245.0 ± 58.9 242.2 ± 60.5 -1.1 0.40\\ 0.60\\\* Lipids (g)^b^ 80.1 ± 34.4 90.0 ± 34.8\* 12.4 0.29\* 0.56\\\* SFA (g)^b^ 27.1 ± 14.0 30.0 ± 12.7 10.7 0.30\\ 0.56\\\* MUFA (g)^b^ 33.9 ± 14.3 39.0 ± 16.3\\ 15.0 0.26\* 0.48\\\* PUFA (g)^b^ 13.1 ± 6.0 14.4 ± 7.0 9.9 0.38\\ 0.46\\\* EPA (g)^b^ 0.06 ± 0.07 0.05 ± 0.04 -16.7 0.33\\ 0.33\\ DHA (g)^b^ 0.17 ± 0.23 0.12 ± 0.08 -29.4 0.30\\ 0.30\\ Trans fat (g)^b^ 3.4 ± 2.0 3.6 ± 2.6 5.9 0.45\\\* 0.56\\\* Cholesterol (mg)^b^ 280.1 ± 139.8 300 ± 116 7.5 0.30\\ 0.36\\ Dietary fiber (g) 21.9 ± 6.4 19.7 ± 5.0\* -10.0 0.32\\ 0.38\\ Soluble fiber (g) 7.3 ± 2.0 6.6 ± 1.7\\ -9.6 0.22 0.27\* Insoluble fiber (g) 14.4 ± 4.7 13.2 ± 3.5 -8.3 0.32\\ 0.37\\ Alchool (g)^b^ 11.4 ± 11.2 10.9 ± 10.7 -4.4 0.61\\\* 0.66\\\* Vitamin C (mg) 138.5 ± 57.3 137.0 ± 62.2 -1.1 0.19 0.19 Folate (mcg) 394.6 ± 115.4 383.2 ± 107.1 -2.9 0.32\\ 0.39\\ Calcium (mg) 898.4 ± 320.5 962.3 ± 399.8 7.1 0.49\\\* 0.56\\\* Iron (mg)^b^ 15.1 ± 4.3 14.1 ± 4.1 -6.6 0.43\\ 0.53\\\* Value are means ± SD ^a^(value derived from FFQ-value derived from 3-day food record)/(value derived from 3-day food record) × 100 ^b^Analyses were performed on log transformed values Significant difference between the two methods \p ≤ 0.05; \\p ≤ 0.01; \\\p ≤ 0.0001 Correlations between FFQ and 3-day food record were statistically significant at \* p ≤ 0.05; \\ p ≤ 0.01; \\\* p ≤ 0.0001 ::: Spearman correlations between values of nutrient intakes measured by the 3-day food record and those assessed by the FFQ at week 0 of the nutritional intervention are shown in Table [1](#T1){ref-type="table"}. Analyses were performed on unadjusted as well as on energy-adjusted values. The average correlation coefficient for the nutrients presented in Table [1](#T1){ref-type="table"} was 0.44 at week 0 (energy-adjusted). Values derived from the FFQ and the 3-day food record were well correlated (energy-adjusted) for CHO, lipids, SFA, trans fatty acids, alcohol, calcium and iron (0.5 \< r \< 0.7) and moderately well correlated for proteins, MUFA, PUFA, cholesterol, dietary fibers, insoluble fibers, EPA and DHA (0.3 \< r \< 0.5). Further adjustment for age and BMI did not materially modify these correlations. Correlations between FFQ and dietary food record at week 12 were slightly higher than correlations at week 0 (data not shown). Table [2](#T2){ref-type="table"} presents changes observed in response to the 12-week nutritional intervention derived from the 3-day food record and from the FFQ. For the majority of the nutrients analyses changes observed by the 3-day food record were similar to those observed by the FFQ except for vitamin C intake for which a higher increase was noted when dietary changes were assessed by the FFQ. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Changes in daily energy and selected nutrients intakes derived from the 3-day food record and from the FFQ in response to the 12-week nutritional intervention (n = 71). ::: Dietary record FFQ ------------------------- -------------------- --------------- Energy (kcal) -197 ± 464 -200 ± 500 Protein (g) -1.4 ± 15.8 -0.5 ± 18.8 CHO (g) -13.9 ± 60.8 -7.4 ± 59.2 Lipids (g) -12.3 ± 32.5 -16.1 ± 28.3 SFA (g) -6.2 ± 13.1 -9.6 ± 9.8 MUFA (g) -4.2 ± 14.7 -4.7 ± 15.3 PUFA (g) -1.1 ± 6.2 -0.8 ± 6.2 EPA (g) 0.06 ± 0.13 0.06 ± 0.06 DHA (g) 0.10 ± 0.40 0.11 ± 0.11 Trans fat (g) -1.4 ± 2.1 -1.6 ± 2.4 Cholesterol (mg) -51.9 ± 142.6 -65.8 ± 92.2 Dietary fiber (g) 3.0 ± 8.0 4.7 ± 6.9 Soluble fiber (g) 0.7 ± 2.7 1.2 ± 2.3 Insoluble fiber (g) 2.3 ± 5.7 3.5 ± 4.8 Alcohol (g) -2.8 ± 10.4 -1.9 ± 9.4 Vitamin C (mg) 3.6 ± 65.9 29.4 ± 62.9\* Folate (mcg) -7.8 ± 131.9 18.1 ± 113.8 Calcium (mg) 3.8 ± 304.1 33.6 ± 331.9 Iron (mg) -0.3 ± 5.3 1.0 ± 4.3 Value are means ± SD Significant difference between the two methods \p ≤ 0.05 ::: Agreement between quartile classification of FFQ and 3-day food record is show in Table [3](#T3){ref-type="table"}. Percentage of agreement varied from 29.4% for proteins to 64.7% for trans fatty acids for the lowest intakes (1^st^tertile) and from 27.8% for CHO to 55.6% for alcohol for the highest intakes (4^th^tertile). When considering all nutrients studied, it was found that, on average, 35.1% of subjects were categorized exactly in the same quartile by the FFQ and by the 3-day food record and 5.1% of the subjects were misclassified in extreme quartiles i.e. subject in the first quartile according to one method and in the fourth quartile according to the other (not shown). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Percentage of agreement for the classification into quartiles of the distribution of selected dietary variables using either the 3-day food record or the FFQ at week 0. ::: Lowest quartile with 3-day food record and FFQ (%)* Highest quartile with 3-day food record and FFQ (%) Exact agreement across quartiles (%) ----------------------- -------------------------------------------------------- --------------------------------------------------------- ------------------------------------------ Energy (kcal) 47.1 38.9 39.4 Protein (g) 29.4 38.9 28.2 CHO (g) 41.2 27.8 26.8 Lipids (g) 35.3 38.9 33.8 SFA (g) 41.2 33.3 29.6 MUFA (g) 35.3 36.8 33.8 PUFA (g) 47.1 38.9 35.2 Trans fat (g) 64.7 38.9 47.9 Cholesterol (mg) 41.2 44.4 36.6 Dietary fiber (g) 47.1 44.4 31.0 Alcohol (g) 52.9 55.6 43.7 ::: Study 2: Reproducibility ------------------------ The 32 subjects in study 2 had a mean BMI of 23.9 ± 3.6 kg/m^2^for women and 27.6 ± 5.1 kg/m^2^for men. The mean age was 42.5 ± 10.4 y for women and 41.2 ± 11.9 y for men. Table [4](#T4){ref-type="table"} shows average daily nutrient intakes derived from the two FFQs (FFQ1, FFQ2) administered 4 weeks apart by two different dietitians. Measurement of total energy intake was not different between the two FFQs (difference of 6.8%). Also, the intakes of proteins, CHO, lipids, SFA, MUFA PUFA, trans fatty acids, cholesterol, alcohol and micronutrients measured by FFQ1 and by FFQ2 did not differ significantly. Adjustment for total energy intake did not alter these observations. Similar results were observed when analyses were performed within each gender separately. Subjects in the first group (in which dietitian \#1 administered the 1^st^FFQ) showed similar variation in nutrients intakes between FFQ1 and FFQ2 than subjects from the 2^nd^group (in which dietitian \#2 administered the first FFQ). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Mean values of daily intakes of nutrients from two FFQ^a^and Spearman correlation coefficients between values derived from the two FFQs^a^. ::: FFQ1 FFQ2 Difference (%)^c^ Unadjusted^d^ Energy-adjusted^d^ ------------------------- ---------------- ---------------- ----------------------- ------------------- ------------------------ Energy (kcal) 2283 ± 584 2128 ± 480 -6.8 0.73\\\* Protein (g)^b^ 89.4 ± 28.4 86.8 ± 24.4 -2.9 0.65\\\* 0.83\\\* CHO (g) 286.3 ± 88.2 262.2 ± 80.6 -8.4 0.79\\\* 0.66\\\* Lipids (g)^b^ 84.6 ± 25.1 78.9 ± 19.8 -6.7 0.47\* 0.82\\\* SFA (g)^b^ 28.8 ± 10.1 27.1 ± 8.6 -5.9 0.51\* 0.81\\\* MUFA (g)^b^ 34.9 ± 11.0 32.6 ± 8.4 -6.6 0.52\* 0.82\\\* PUFA (g)^b^ 14.3 ± 6.2 13.1 ± 4.7 -8.4 0.60\\ 0.74\\\* EPA (g)^b^ 0.06 ± 0.05 0.07 ± 0.06 16.7 0.55\\ 0.55\\ DHA (g)^b^ 0.12 ± 0.09 0.15 ± 0.12 25.0 0.55\\ 0.56\\ Trans fat (g) 3.5 ± 1.7 3.2 ± 1.4 -8.6 0.60\\ 0.79\\\* Cholesterol (mg) 278.2 ± 102.8 269.3 ± 75.1 -3.2 0.47\* 0.73\\\* Dietary fiber (g) 22.9 ± 7.9 20.8 ± 6.3 -9.2 0.76\\\* 0.75\\\* Soluble fiber (g) 7.8 ± 2.4 7.1 ± 2.0 -9.0 0.82\\\* 0.87\\\* Insoluble fiber (g) 15.0 ± 5.5 13.7 ± 4.4 -8.7 0.73\\\* 0.81\\\* Alchool (g) 9.6 ± 6.7 9.4 ± 8.0 -2.1 0.76\\\* 0.70\\\* Vitamin C (mg) 201.6 ± 103.5 165.2 ± 77.0 -18.1 0.81\\\* 0.62\\\* Folate (mcg)^b^ 442.0 ± 147.4 394.5 ± 128.5 -10.7 0.71\\\* 0.69\\\* Calcium (mg)^b^ 1153.5 ± 453.5 1085.1 ± 456.2 -5.9 0.79\\\* 0.86\\\* Iron (mg)^b^ 15.8 ± 5.7 14.5 ± 4.7 -8.2 0.64\\\* 0.79\\\* Values are means ± SD ^a^2 FFQ were administered 4 weeks apart ^b^Student t test analyses were performed on log transformed values ^c^(value derived from FFQ2 - value derived from FFQ1)/(value derived from FFQ1) × 100 ^d^Correlations between FFQ1 and FFQ2 were statistically significant at \* p ≤ 0.05; \\ p ≤ 0.01; \\\* p ≤ 0.0001 ::: Table [4](#T4){ref-type="table"} presents Spearman correlations between nutrient intakes measured by FFQ1 and FFQ2. The average correlation coefficient for the nutrients presented in Table [4](#T4){ref-type="table"} is 0.74. Values derived from the two FFQs after adjustment for energy intake were generally very well correlated for proteins, lipids, SFA, MUFA, PUFA, trans fatty acids, cholesterol, dietary fibers, soluble and insoluble dietary fibers, alcohol, calcium and iron (0.7 \< r \< 0.9) and well correlated for CHO, EPA, DHA, vitamin C and folate (0.6 \< r \< 0.7). Partial correlations between nutrients derived from FFQ1 and those derived from FFQ2 were unchanged after adjusting for age, BMI and for the dietitian who administered the 1^st^FFQ. In addition, similar correlations were observed when analyses were computed within each gender separately (not shown). Table [5](#T5){ref-type="table"} shows agreement between quartile classification when FFQ1 was compared to FFQ2. Percentage of agreement varied from 28.6% for proteins to 75.0% for energy and alcohol for the lowest intakes (1^st^quartile) and from 37.5% for SFA and MUFA to 75.0% for energy, proteins, CHO and alcohol for the highest intakes (4^th^quartile). When considering all the nutrients studied, it was found that on average, 45.7% of subjects were categorized exactly in the same quartile by both FFQs and 2.6% of the subjects were misclassified in extreme quartile (not shown). ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Percentage of agreement for the classification into quartiles of the distribution of selected dietary variables using either FFQ1 or FFQ2. ::: Lowest quartile with FFQ1 and FFQ2 (%) Highest quartile with FFQ1 and FFQ2 (%) Exact agreement across quartiles (%) ----------------------- -------------------------------------------- --------------------------------------------- ------------------------------------------ Energy (kcal) 75.0 75.0 62.5 Protein (g) 28.6 75.0 37.5 CHO (g) 50.0 75.0 59.4 Lipids (g) 50.0 62.5 40.6 SFA (g) 50.0 37.5 34.4 MUFA (g) 62.5 37.5 40.6 PUFA (g) 50.0 62.5 46.7 Trans fat (g) 62.5 50.0 40.6 Cholesterol (mg) 37.5 50.0 40.6 Dietary fiber (g) 62.5 50.0 43.8 Alcohol (g) 75.0 75.0 56.3 ::: Discussion ========== Accurate assessment of dietary intakes and dietary changes plays a central role in nutritional studies. Each tool used to evaluate dietary intakes has some strengths and limitations. Also, all standard dietary assessment methods are subjected to bias such as underreporting \[[@B6]\]. In our study, we developed a 91-items interviewer-administered FFQ that was sufficiently accurate to measure intakes of nutrients in the habitual diet of subjects from the Québec City metropolitan area and changes in nutrient intakes following a 12 week intervention promoting the Mediterranean food pattern. In the validation study, coefficients of correlation between values derived from the FFQ and those obtained by the 3-day food record ranged from 0.30 to 0.60 for macronutrients and from 0.19 to 0.56 for micronutrients at week 0. It has been previously reported that correlation coefficients for validation studies ranged from 0.4 to 0.7, similar to our results after energy adjustment \[[@B6],[@B15]\]. Also, our interviewer-administered FFQ did not significantly overestimate energy intake compared to a 3-day food record. The fact that the interviewer used food models to facilitate the estimation of portion size can contribute to explain this findings. It has been shown that FFQ can both under- and overestimate intakes of specific nutrients. In fact, many validation studies have reported that FFQ, as compared to food-record or 24-hour recall overestimate nutrients intakes as well as energy intake \[[@B16]-[@B20]\]. In contrast, other studies have reported that FFQ did not systematically overestimate energy and nutrients intakes \[[@B14],[@B21]-[@B23]\]. Despite the fact that we obtained similar values for energy intake with both dietary methods, we can not exclude the possibility that both tools are subjected to underreporting and therefore underestimate usual dietary intakes. It has been previously suggested that subjects may tend to underreport actual food intake by as much as 20% when completing a weighted dietary record \[[@B24]\]. It has been argued that subjects who complete 3-day food record may change their nutritional food habits in order to simplify the recording of food intakes or to impress the dietitian. Also, errors in 3-day food records can be attributable to interpretation of the dietitian encoding the records. In our study, the same dietitian verified all the food records to make sure that dietary data were coded similarly for all participants. In the present study, 38% of subjects included in the validation study at week 0 had a ratio between energy intake to estimate BMR below 1.35. Considering that they had to be weight stable to be included in the study it is likely that these women were underestimating their habitual diet. Black et al concluded in a review that underreporting was observed in a great majority of nutritional surveys independently of the method used \[[@B25]\]. Earlier studies conducted in lean women demonstrated that underreporting was mainly explained by undereating \[[@B26]\] or underreporting snack foods \[[@B27]\] whereas in obese subjects underreporting could be explained by an underestimation in recording portion size and to social desirability. In addition, underreporting occurs more often among foods considered \'bad\' or \'unhealthy\' \[[@B28]\]. In our validation study, there were no significant differences between BMI of women who were considered as underreporters and women who did not underreport (not shown). In a nutritional intervention, interpretation of the study outcomes with regard to dietary changes will depend not only of the validity and the reproducibility of the method used but also of the sensibility of the method to detect dietary changes in response to the intervention. In our nutritional intervention study, conducted in a sample of healthy women, both diet assessment methods detected similar dietary changes over the duration of the intervention. These findings suggested that our FFQ is sensitive to dietary changes in response to our intervention and could be used to assess dietary changes during a nutritional intervention. Our results are in agreement with study that showed that in response to a nutritional intervention a FFQ measured similar dietary changes as compared to 24-hour recalls \[[@B29]\] or 4-day food records \[[@B7]\]. The major differences between the two methods in our study were noted for total lipids and MUFA intakes. Our FFQ was designed to assess precisely lipid intake and many questions were asked about types of fat used to spread or to cook. The more important differences between FFQ and 3-day food record for MUFA and lipids could therefore be explained by the fact that it was difficult for participants to report precisely their lipid consumption when completing the FFQ. It has also been reported in obese men that underreporting of food record is usually specific to lipid intake \[[@B30]\] and it is thus possible that some women did not record all fats or foods high in fat consumed when completing their 3-day food record. Therefore, it is difficult to determine whether our FFQ tended to overestimate lipid or whether the 3-day food record tended to underestimate it. Also, dietary changes for these nutrients were in the same magnitude in response to our intervention with both methods. On the other hand, Mediterranean diet is usually considered high in MUFA. In North America, MUFA are mostly provided by partially hydrogenated vegetable oils and animals products \[[@B31]\]. In that context, MUFA to SFA ratio could be considered as a better indicator of a Mediterranean diet. In our study, we noted that this ratio was not different between the two methods at baseline and changes observed in response to the nutritional intervention did not differ significantly (not shown). The agreement in quartile classification was acceptable for selected nutrients with a mean of 35.1% of subjects who were in exact agreement and 5.1% who were misclassified in extreme quartiles. This finding is similar to previous observations \[[@B16],[@B19],[@B32],[@B33]\]. In many studies, classification in the same segment of the distribution using two different methods is found in 30% to 40% of subjects \[[@B16],[@B32],[@B34]\]. When analyses were performed at week 6 and 12 after the beginning of the nutritional intervention, coefficients of correlation were slightly higher than at week 0. We suggest that this finding be partly explained by the intervention effect. As previously reported \[[@B35]\] subjects could be influenced by a learning effect. In fact, subjects could be influenced by the first FFQ experience and be more adequately prepared for the second FFQ. The nutritional intervention may have also influenced the manner in which subjects were completing their 3-day food records during the study. In a nutritional intervention, it is important to use a reproducible method to insure that dietary changes observed are due to the intervention effects and not to the instrument error. Our study suggests that the FFQ presents a good degree of reproducibility. In fact, in reproducibility studies the coefficients of correlation generally ranged from 0.5 to 0.7 \[[@B6]\]. In our study, coefficients of correlation ranged, after energy adjustment, from 0.62 for vitamin C to 0.83 for protein intakes. These values are similar to correlations reported by others \[[@B14],[@B18],[@B19],[@B21],[@B22],[@B33],[@B35]-[@B37]\]. In our reproducibility study, lower mean energy intake and nutrient intakes were found at the second administration of the FFQ as compared to the first FFQ (difference of approximately 10%). However, relatively high and uniform correlation coefficients for values derived from the two FFQs were observed. Riley et al \[[@B35]\] also reported with an administered FFQ that energy intake was 10% lower at the second FFQ administration and this reduction was uniform for all nutrients studied. In our study, intakes of most nutrients were systematically higher when measured with the first FFQ compared to the second one, except for alcohol consumption, which remained the same. Seasonal variation can not explain this difference because both FFQs were administered during the same season. The fact that subjects estimated a lower frequency of intake during the second administration of the FFQ may be explained by their earlier experience in completing the FFQ. Better general knowledge of dietary intakes could lead to a readjustment in estimation of intakes after the first administration of the FFQ and therefore changes in estimated energy intake. Conclusions =========== In conclusion, the FFQ that we developed to estimate usual average daily energy and nutrient intakes in subjects from the Québec City metropolitan area is valid and reproducible. Mean energy and nutrient intakes were estimated accurately by our FFQ compared to the 3-day food record. The fact that our FFQ was administered by a dietitian trained to insure a standardized administration of FFQ was important to optimize the validity and reproducibility of the method. We also showed that both methods appeared to underestimate energy intake in a large proportion of subjects. Usually, food records are considered as the gold standard method to assess dietary intakes. It is however important to recognize that food records also have their own limitations. In nutritional studies, an interviewer-administered FFQ, as the one we used in the present study, can be used to assess energy and nutrient intakes and requires less time to compute dietary informations than food records. FFQ also decreases the possibility of interpretation by the coding person. Finally, there is still a need to develop other efficient methods to measure dietary intakes that will permit to more closely match habitual dietary intakes of individuals in their living environment. List of abbreviations used ========================== CHD: coronary heart disease FFQ: Food Frequency Questionnaire BMR: basal metabolic rate (BMR) BMI: body mass index NDS-R: Nutrition Data System for Research CHO: carbohydrates SFA: saturated fatty acids MUFA: monounsaturated fatty acids PUFA: polyunsaturated fatty acids EPA: eicosapentaenoic acid DHA: docosahexaenoic acid Competing interests =================== None declared. Authors\' contributions ======================= JG participated to data collection, performed data analysis and drafted the manuscript. GN and AL participated to data collection. BL and SL conceived the study, and participated in its design and coordination. All authors read and approved the final version of the manuscript. Acknowledgements ================ S.L. is a research scholar from the Fonds de la recherche en santé du Québec and B.L is the recipient of a Canada Research Chair in Nutrition, Functional Foods and Cardiovascular Health from the Canada Research Chair Program. This study was partly supported by the Canada Research Chair in Nutrition, Functional Foods and Cardiovascular Health from the Canada Research Chair Program. The authors express their gratitude to the subjects and their family for their motivation and implication throughout the study. We acknowledge the contribution of Nancy Gilbert R.D, M.Sc, and Amélie Charest R.D, for the nutritional intervention.	PubMed Central	2024-06-05T03:55:48.463812	2004-9-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522826/", "journal": "Nutr J. 2004 Sep 13; 3:13", "authors": [ { "first": "Julie", "last": "Goulet" }, { "first": "Geneviève", "last": "Nadeau" }, { "first": "Annie", "last": "Lapointe" }, { "first": "Benoît", "last": "Lamarche" }, { "first": "Simone", "last": "Lemieux" } ] }
PMC522868	Background ========== The transcription of protein-coding genes in distinct temporal and spatial patterns plays a central role in the differentiation and development of animal embryos. Decoding how the unique expression pattern of every transcript is encoded in DNA is essential to understanding how genome sequences specify organismal form and function. Understanding gene regulation requires discovering the cis-acting sequences that control transcription, identifying which trans-acting factors act on each regulatory sequence, and determining how these interactions affect the timing and organization of transcription. The first step in this process is by no means straightforward. Regulatory regions are often large and complex. Functional cis-acting sequences are found 5\' and 3\' of transcripts and in introns, and can act over short or long distances. Most of the described animal regulatory sequences were identified by experimental dissection of a locus, and astonishingly few of these are well characterized. Despite the paucity of good examples, as multiple regulatory sequences from different organisms were identified and characterized, some common features became apparent \[[@B1],[@B2]\]. Most animal regulatory sequences act as compact modular units, with regions of roughly a kilobase (kb) in size controlling specific aspects of a gene\'s transcription. These regulatory units - referred to here as cis-regulatory modules (CRMs) - tend to contain functional binding sites for several different transcription factors, often with multiple sites for each factor. As the first animal genome sequences were completed \[[@B3]-[@B6]\], researchers began to tackle the challenge of identifying regulatory sequences on a genomic scale. We and several other groups began to ask whether common characteristics of regulatory sequences - modularity and high binding-site density - might be distinguishing characteristics that would permit the computational identification of new regulatory sequences. A number of in silicomethods to identify regulatory sequences on the basis of binding-site clustering have been developed and applied to animal genomes \[[@B7]-[@B10]\]. Some of the predictions have the expected in vivoregulatory activity \[[@B11]-[@B17]\], yet few of these predictions have been systematically evaluated. The transcriptional regulatory network governing early Drosophiladevelopment is perhaps the best system in which to apply and evaluate these methods. Development of the Drosophilaembryo is arguably better understood than that of any other animal. Sophisticated genetic screens \[[@B18],[@B19]\] have identified most of the key regulators of early development, and the molecular biology and biochemistry of these factors and their target sequences have received a great deal of attention. The spatial and temporal embryonic expression patterns of a large number of genes are known from microarray \[[@B20]\] and in situexpression studies \[[@B21]\]. Transcriptional regulation plays a uniquely important role in pre-gastrula patterning, as most of the key events occur in the absence of cell membranes and the cell-cell signaling systems that play a crucial role later in fly development and throughout the development of most other animals. In a previous study \[[@B11]\], we identified 37 regions of the Drosophila melanogastergenome with unusually high densities of predicted binding sites for the early-acting transcription factors Bicoid (BCD), Hunchback (HB), Krüppel (KR), Knirps (KNI) and Caudal (CAD). As nine of these regions overlapped previously known CRMs, we proposed the remaining 28 as predicted CRMs (pCRMs). We tested one of the previously untested pCRMs for enhancer activity in a standard reporter gene assay \[[@B22],[@B23]\] and showed that it is responsible for directing a portion of the embryonic expression pattern of the gap transcription factor gene giant(gt) in a posterior stripe. Here, we report the systematic testing of the remaining 27 untested pCRMs for enhancer activity, resulting in collections of both bona fidepositive and false-positive predictions, allowing us to develop and evaluate methods to improve the accuracy of methods for identifying functional cis-regulatory sequences. We were particularly interested in methods based on the comparison of genome sequences of related species. The genome sequence of D. pseudoobscura(which diverged from D. melanogasterapproximately 46 million years ago \[[@B24]\]) was recently completed by the Baylor Human Genome Sequencing Center, and several other Drosophilaspecies are currently being sequenced. The morphological and molecular events in early embryonic development are highly conserved among drosophilids, and we expect the activity of the transcriptional regulators and the architecture of regulatory networks to be highly conserved as well. Most D. melanogasterregulatory sequences should have functional orthologs in other Drosophilaspecies \[[@B25],[@B26]\], and a major rationale for sequencing other Drosophilaspecies is the expectation that regulatory sequences have characteristic patterns of evolution that can be used to identify them and to better understand their function. Most methods used to identify regulatory sequences from interspecies sequence comparison are fairly simple. They identify \'conserved\' non-coding sequences (CNSs), operationally defined as islands of non-coding sequence with relatively high conservation flanked by regions of low conservation, and assume that this conservation reflects regulatory function. Although crude, these methods have been remarkably effective in identifying mammalian regulatory sequences \[[@B27],[@B28]\], and preliminary studies in Drosophilasuggest that similar methods will be valuable in insects as well \[[@B29]\]. However, despite such successes, the extent of the efficacy of comparative sequence analysis in regulatory sequence discovery remains unclear. A systematic comparison of human-mouse sequence conservation in known regulatory regions and ancestral repeats (which provide a model for neutral evolution) suggests that regulatory regions cannot generally be distinguished on the basis of simple sequence conservation measures alone \[[@B30],[@B31]\]. Similarly, a recent analysis of D. melanogasterand D. pseudoobscurashowed that known regulatory regions are only slightly more conserved than the rest of the non-coding genome \[[@B32]\], highlighting the need for further study and the development of comparative methods that go beyond measures of sequence identity. Results ======= Expression patterns of pCRM containing transgenes ------------------------------------------------- The 37 pCRMs are shown in Table [1](#T1){ref-type="table"}. Each has been assigned an identifier (of the form PCEXXXX). The first nine overlap previously known enhancers of runt(run), even-skipped(eve), hairy(h), knirps(kni) and hunchback(hb). To determine whether any of the remaining 28 pCRMs also function as enhancers, we generated P-element constructs containing the pCRM sequence with minimal flanking sequence on both sides fused to the evebasal promoter and a lacZreporter gene (see Materials and methods). As the margins of the tested sequences do not precisely correspond to the margins of the clusters, we assigned a unique identifier (of the form CEXXXX) to each tested fragment (identical CE and PCE numbers correspond to the same pCRM). We successfully generated multiple independent transgenic fly lines for 27 of the 28 pCRMs. We repeatedly failed to generate transgenes containing CE8007. This sequence contains five copies of an approximately 358 base-pair (bp) degenerate repeat. One additional pCRM (CE8002) also contains tandem repeats. While we were able to generate transgenes for CE8002 and assay its expression, these two tandem repeat-containing pCRMs (CE8007 and CE8002) were excluded from subsequent analyses. We examined the expression of these constructs by in situRNA hybridization to the lacZtranscript in embryos at different stages in at least three independent transformant lines. Nine of the 27 transgenes showed mRNA expression during embryogenesis (Figure [1](#F1){ref-type="fig"}), while the remaining 18 assayed transgenes showed no detectable expression at any stage during embryogenesis. To identify the genes regulated by the nine pCRMs with embryonic expression, we examined the expression patterns of genes containing the pCRM in an intron and genes with promoters within 20 kb of the CRM (see Figure [1](#F1){ref-type="fig"}). We used the embryonic microrarray and whole-mount in situexpression data available in the Berkeley Gene Expression Database \[[@B21]\], supplemented with additional whole-mount in situexperiments where necessary (data not shown; these new in situ\'s will be included in the public expression database \[[@B33]\] at its next release). Six of the active pCRMs drive lacZexpression in patterns that recapitulate portions of the expression of a gene adjacent to or containing the pCRM. Four of these new enhancers act in the blastoderm and two during germ-band elongation. CE8001 is 5\' of the gene for the gap transcription factor giantand recapitulates the posterior domain (65-85% egg length measuring from the anterior end of the embryo) of gtexpression in the blastoderm as previously described \[[@B11]\]. CE8011 is 5\' of the gene for the POU-homeobox transcription factor nubbin (nub). The CRM recapitulates the endogenous blastoderm expression pattern of nub, first detected as a broad band extending from 50 to 75% egg length. Although nubexpression continues in later embryonic stages, CE8011 expression is limited to the blastoderm stage. CE8010 is 5\' of the pair-rule gene odd-skipped(odd) and drives expression of two of its seven stripes: stripe 3 at 55% and stripe 6 at 75% egg length. This CRM also has the ability to drive later, more complex, patterns of expression. During stages 6 and 7, expression is detected in the procephalic ectoderm anlage and in the primordium of the posterior midgut. By stage 13, expression is also detected in the anterior cells of the midgut which will give rise to the proventriulus, the first midgut constriction, the posterior midgut and microtubule primordial as well as cells in the hindgut, all similar to portions of the pattern of wildtype oddprotein expression previously described \[[@B34]\]. CE8024 is 3\' of the pair-rule gene fushi-tarazu(ftz) and drives expression of two of its stripes: stripe 1 at 35% and stripe 5 at 65% egg length. Using a similar CRM reporter assay, this pattern of expression was also detected by \[[@B35]\]. CE8012 is in the third intron of POU domain protein 2(pdm2) and appears to completely recapitulate its stage-12 expression pattern, which is limited to a subset of the developing neuroblasts and ganglion mother cells of the developing central nervous system. A similar pattern of expression was previously described for the protein product of pdm2\[[@B36]\]. It is worth noting that we do not detect expression of CE8012 in the blastoderm stage, whereas the endogenous gene exhibits a blastoderm expression pattern similar to nub. CE8027 is 3\' of the gene for the Zn-finger transcription factor squeeze (sqz) and recapitulates the wild-type expression pattern of sqzRNA in a subset of cells in the neuroectoderm at stage 12. The wild-type sqzexpression pattern was previously described \[[@B37]\]. The remaining three active pCRMs cannot be easily associated with a specific gene. CE8005 drives expression in the ventral region of the embryo. It is 3\' of a gene encoding a ubiquitously expressed Zn-finger containing protein (CG9650) that is maternally expressed and deposited in the embryo. This strong maternal expression potentially obscures a zygotic expression pattern. Two additional adjacent genes, CG32725and CG1958, showed no expression in whole-mount in situhybridization of embryos. CE8016 drives a seven-stripe expression pattern in the blastoderm. It is in the first intron of CG14502which shows very low level expression by microarrays in the blastoderm, and has no obvious detectable pattern of expression in whole-mount in situhybridization of embryos. This pCRM is approximately 2 kb 5\' of scribbler (sbb), which is expressed maternally, possibly obscuring an early zygotic expression pattern (a few in situimages show a hint of striping). sbbis also expressed later in development in the ventral nervous system. An additional potential target, Otefin(Ote), is also expressed maternally and relatively ubiquitously through germ-band extension. All other nearby genes displayed in Figure [1](#F1){ref-type="fig"} showed no embryonic expression in whole-mount in situhybridization or by microarray. CE8020 drives an atypical four-stripe pattern in the blastoderm - two stripes at 7% and 26% that are anterior to the first ftzstripe and two stripes at 39% and 87%. It is in the first intron of ome(CG32145), which is not expressed maternally and has no blastoderm expression, but is expressed late in salivary gland, trachea, hindgut and a subset of the epidermis. All other nearby genes displayed in Figure [1](#F1){ref-type="fig"} showed no embryonic expression in whole-mount in situhybridization or by microarray. With these results, and the nine previously known enhancers, at least 15 of the 37 highest density clusters of the five transcription factors used in our initial screen have early-embryonic enhancer activity. The remainder of this paper examines 35 of the original 37 clusters, with the two tandem repeat-containing clusters excluded. We divide these 35 into three categories - 15 positives (the nine overlapping previously known enhancers plus the six new enhancers identified here), three ambiguous (the three positives without a clear regulated gene), and 17 negatives (see Table [2](#T2){ref-type="table"}). We largely focus on differences between the positives and negatives. Distinguishing active and inactive clusters ------------------------------------------- All 15 positives are within 20 kb of the transcription start site (or, where the transcription start site is unknown, the start of the gene annotation) of transcripts expressed in spatiotemporal patterns consistent with regulation by the maternal and gap transcription factors used in our screen (that is, in anterior-posterior patterns in the blastoderm or in the developing neuroblasts of the central nervous system). Only one of the 17 negatives was located within 20 kb of a plausible target (PCE8021 is 7 kb upstream of reaper), so out of 16 pCRMs located within 20 kb of a gene with appropriate expression, 15 (94%) are active enhancers. The positives are, on average, larger than the negatives (average cluster size of positive = 900 bp, while average cluster size of negatives was 711 bp), a difference that is significant by the Komogorov-Smirnov (KS) test (p= 0.017). The positives have a slightly higher density of binding sites, but this difference was not significant. The binding site composition of the positives and negatives are similar (the positives contain more KR, and fewer BCD binding sites, but again these differences are not highly significant). Although others have reported that some factors have characteristic spacings with respect to themselves and other factors \[[@B38]\], we could not find evidence for such spacing or identify other differences that could distinguish positive pCRMs from negative (Figure [2](#F2){ref-type="fig"}). Use of D. pseudoobscura ------------------------- We assembled the D. pseudoobscuragenome from traces deposited in the NCBI\'s TraceDB using the Celera assembler \[[@B39],[@B40]\]. These assemblies were used to examine the conservation of our pCRMs and to assess whether conservation could be used instead of or in addition to binding site clustering as a way to identify CRMs. We first assessed whether positive pCRMs could be distinguished from their flanking sequences based on degree of conservation. In vertebrate comparative genomics, relatively simple methods (such as VISTA \[[@B41]\]) are commonly used to identify CNSs that are a surprisingly rich source of new cis-regulatory sequences. We evaluated the potential of using such methods with D. melanogasterand D. pseudoobscurain two ways. First, we constructed percent-identity plots for the regions containing all of the 37 pCRMs (Figure [3](#F3){ref-type="fig"}; similar plots for all pCRMs are available in the online supplement at \[[@B42]\]) with the location of pCRMs and other known regulatory sequences clearly indicated. Although it appears that some CRMs (that is, evestripe 3/7) would have been successfully identified by such simple comparative methods, positive pCRMs do not collectively appear distinguishable from flanking sequence on the basis of conservation alone. Although positive pCRMs are almost all in highly conserved blocks, there is a surprisingly high amount of non-coding sequence conservation throughout these regions, and most negative pCRMs are also contained in highly conserved blocks. It remains to be seen whether this difference in the conservation landscape of Drosophilanon-coding sequences compared to vertebrates reflects a significant difference in the functional organization of non-coding sequences, or simply indicates that there is too little divergence between D. melanogasterand D. pseudoobscurato detect useful differences in the rates of evolution (see Discussion). We next assessed whether positive pCRMs can be distinguished from negative pCRMs on the basis of their degree of similarity between D. melanogasterand D. pseudoobscura. For each pCRM-containing region, we identified orthologous contigs from the D. pseudoobscuraassembly and aligned them using the alignment program LAGAN \[[@B43]\]. We were able to find orthologous regions for 32 pCRMs (see Table [2](#T2){ref-type="table"}). Using the simple measure of percent identity, we find that positive pCRMs are, on average, more highly conserved than negative pCRMs (see Table [2](#T2){ref-type="table"}). Although this difference is significant (p= 0.002 by KS test), the distribution of conservation scores for positive and negative pCRMs overlap considerably, and thus conservation alone is not a useful way of distinguishing positive and negative pCRMs (see Figure [4b](#F4){ref-type="fig"}). To get a genome-wide perspective on the degree of conservation in positive pCRMs, we analyzed the conservation of CRM-sized (1 kb) regions in randomly chosen sections of the genome (Figure [4b](#F4){ref-type="fig"}). Positive pCRMs are, generally, more conserved than average CRM-sized sequences, and some positive pCRMs are among the most highly conserved non-coding sequences in the genome. However, a conservation cut-off necessary to select the majority of positive pCRMs would select roughly one third of the non-coding regions of the genome, and thus is not a practical method for prioritizing sequences for functional analysis. Conservation of binding sites and conservation of clustering ------------------------------------------------------------ We expect that most genes will have similar expression patterns in D. melanogasterand D. pseudoobscura, and that most D. melanogasterenhancers should have functional orthologs in D. pseudoobscura. For those enhancers we seek to identify here - namely those where binding site clustering reflects their function - we expect clustering to be found in both D. melanogasterand D. pseudoobscura. Conversely, clusters that simply occur by chance in either genome but do not reflect the function of the sequence (as, we believe, is the case for many of our false-positive predictions) should not be conserved. Thus, looking for conservation of binding-site clustering should provide a valuable way of distinguishing functional and non-functional binding-site clusters in the D. melanogastergenome. We used the alignments described above to examine the conservation of individual predicted binding sites in all of the pCRMs (Table [2](#T2){ref-type="table"}). We refer to a predicted D. melanogasterbinding site that overlaps a predicted D. pseudoobscurabinding site for the same factor in an alignment as an \'aligned\' site. We require overlap and not perfect alignment to compensate for alignment ambiguity; the overwhelming majority (85%) of aligned sites are perfectly aligned. Although there is only a subtle difference in the binding-site density in the positive and negative pCRMs in D. melanogaster(22.7 sites/kb compared to 22.2), the density of aligned binding sites in positive pCRMs (13.8 sites/kb) is nearly twice that in negative pCRMs (6.8 sites/kb). This is a highly significant difference (p\< 0.001 by KS test) and aligned site density better discriminates positive and negative pCRMs than sequence conservation (compare Figure [4c](#F4){ref-type="fig"} and [4b](#F4){ref-type="fig"}). Sixty-one percent of the predicted binding sites in positive pCRMs are aligned, while only 30% of the sites in negative pCRMs are aligned. Across the genome, 22.3% of predicted binding sites are aligned meaning that there is a roughly fourfold increase over background in the probability that a binding site in a positive pCRM is conserved in place compared to a binding site in a negative pCRM. Sixty-one percent is almost certainly an underestimate of the fraction of pCRM sites that are functionally conserved. The D. melanogaster-D. pseudoobscuraalignments were not always unambiguous (using simulations we have assessed the role of alignment algorithms in identifying conserved transcription factor binding sites, see \[[@B44]\]), and some orthologous binding sites may not have been properly aligned. More important, studies of the evolution of various Drosophilaenhancers suggest that the positions of binding sites within an enhancer are somewhat plastic, and the functional conservation of a binding site does not necessarily require positional conservation \[[@B25],[@B26]\]. To characterize the extent of binding site conservation independent of positional conservation, we computed a second measure of binding-site conservation. We consider an unaligned binding site in D. melanogasterto be \'preserved\' if it can be matched to a corresponding site in the D. pseudoobscurapCRM (allowing each D. pseudoobscurasite to match only one D. melanogaster site). If we consider both aligned and preserved sites to be conserved, then roughly 80% of the sites in positive pCRMs are conserved compared with 40% in negative pCRMs. The density of preserved but not aligned sites in positive pCRMs (4.3/kb) is considerably higher than in negative pCRMs (2.2/kb) or random sequences (1.8/kb). Thus, in the D. pseudoobscuraorthologs of active D. melanogasterCRMs we observe an increase in binding-site density that cannot be explained by the positional conservation of sites found in D. melanogasteror the random occurrence of sites in the genome. Several of the 15 positive CRMs have high densities of these preserved but unaligned sites, but two in particular, runtstripe 3 and hairystripe 6, stand out from the rest. These two have almost as many preserved sites as strictly aligned sites. Aligned plus preserved (conserved) site density (Figure [4d](#F4){ref-type="fig"}) almost perfectly separates positive from negative pCRMs. Only one of the positive pCRMs (PCE8012) has a conserved site density below 14 sites/kb, while only one of the negative pCRMs (PCE8021) has a conserved site density above 14 sites/kb. eCIS-ANALYST: a comparative enhancer finder ------------------------------------------- As the conservation of binding sites and binding-site clusters between D. melanogasterand D. pseudoobscurasuccessfully distinguishes positive and negative predictions made using the D. melanogastersequence alone, we incorporated comparative sequence data into our enhancer-prediction algorithm CIS-ANALYST \[[@B11]\]. Instead of searching for clusters of predicted binding sites in a single genome, eCIS-ANALYST (the \'e\' is for evolutionary) searches for conserved clusters of sites between the two genomes (see Materials and methods). eCIS-ANALYST is available at \[[@B45]\]. Using 17 negative pCRMs and an expanded set of 25 positive pCRMs (which included the 15 positive predictions discussed above and 10 functional enhancers known to respond to the five factors; these 10 additional enhancers were discussed and analyzed in \[[@B11]\] but had binding-site densities below the threshold used there), we compared the ability of CIS-ANALYST and eCIS-ANALYST to identify positive pCRMs and to distinguish positive and negative pCRMs at different binding-site density cutoffs (Figure [5](#F5){ref-type="fig"}). The incorporation of the conservation criteria greatly improves the algorithm\'s apparent performance. The expected fraction of false positives is markedly reduced, and it is possible to lower the binding site threshold to recover six of the ten previously missed positive enhancers without increasing the number of expected false-positive predictions. New predictions --------------- As eCIS-ANALYST has markedly better specificity than CIS-ANALYST, we sought to identify BCD, HB, KR, KNI and CAD targets that were missed with the relatively stringent criteria used in our previous analysis. Rather than use a stringent cutoff (15 binding sites per 700 bp) as we did in \[[@B11]\], we performed three separate runs with lower cutoffs (for example, 10 sites per 700 bp in one run) and applied a conservation threshold (see Materials and methods and Additional data file 3) to select 929 conserved binding-site clusters. There were 842 new pCRMs within 20 kb or in an intron of an annotated transcript (Additional data file 7) and 87 more than 20 kb (Additional data file 8). We ranked these new pCRMs by a simple scoring scheme that measures both the density and the total number of sites conserved (we evaluated several different scoring schemes, and selected one that optimally identified regions near genes with blastoderm expression patterns; see Materials and methods). The 75 highest-scoring pCRMs within 20 kb of an annotated transcript are shown in Table [3](#T3){ref-type="table"}. Thirteen of the 15 positive pCRMs described above are in the top 75 (ftzstripe 1/5 is number 107 and the pdm2neurogenic enhancer is number 418) as are five other known enhancers. One of our negative pCRMs, CE8021, is ranked number 12. To focus our search for new enhancers on genes likely to be regulated by BCD, HB, KR, KNI and/or CAD, we searched FlyBase \[[@B46]\] and a database of Drosophilaembryonic expression patterns \[[@B21]\] and identified 278 genes with anterior-posterior patterns in the blastoderm (AP genes; Figure [6](#F6){ref-type="fig"} and see also Additional data files 2 and 9). Thirty-one of the 75 highest-scoring new predictions are adjacent to or within 20 kb of one or more of these genes, including 11 pCRMs that do not overlap previously described enhancers. The 75 highest-scoring predictions within 20 kb of an AP gene but not in Table [3](#T3){ref-type="table"}, are shown in Table [4](#T4){ref-type="table"}. In Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"} together, there are 106 high-scoring conserved binding-site clusters near AP genes, 90 of which do not overlap known enhancers. Discussion ========== We performed a large and comprehensive evaluation of the efficacy of computational methods for the identification of functional cis-regulatory modules in Drosophila. Analysis of the in vivoactivity of 36 high-density clusters of predicted BCD, HB, KR, KNI and CAD binding sites identified in our previous study \[[@B11]\] offers compelling support for the use of transcription factor binding-site clustering as a method to identify regulatory sequences, as at least 15 of these sequences function as early developmental enhancers in vivo. An evolutionary analysis of these sequences - based on comparisons of the D. melanogasterand D. pseudoobscuragenomes - shows that sequence conservation alone can not reliably discriminate cluster-containing regions that function in vivofrom those that do not. However, a new method that combines binding-site clustering and comparative sequence analysis to search for binding-site clusters that are present in multiple species does reliably discriminate active and inactive clusters. Using this method, we make several hundred predictions of new CRMs, a large number of which are located near likely target genes. Binding-site clustering ----------------------- The success of relatively simple binding-site clustering methods here and in other work is remarkable given the crudeness of these methods. As our negative predictions demonstrate, the mere presence of a cluster of binding sites is not sufficient to make an active embryonically expressed CRM. Although these 17 sequences have binding-site densities and compositions indistinguishable from their functional cousins, they do not function as enhancers in a simple transgene assay. It is possible that some of these negative pCRMs may be functional enhancers that respond to the factors used in our screen, perhaps requiring a different promoter or other flanking sequences not used in the transgene. While further experiments could address this possibility, we felt these were a low priority, as few of the D. pseudoobscuraorthologs of these negative pCRMs have binding-site clusters, and few are near genes with appropriate expression patterns. Thus it is unlikely that many function in their endogenous locations in vivo. Both the general activity and, more important, the specific regulatory output of a CRM are a complex, and still poorly understood, function of the specific architecture of its sites. The emerging picture of the ordered multiprotein complexes that mediate enhancer activity suggests constraints on enhancer composition and architecture \[[@B1],[@B2],[@B47]\] whose elucidation will form a critical part of the future dissection of the function of cis-regulatory sequences. It is intriguing that three of the clusters we tested direct expression patterns that bear no obvious relationship to the expression of a neighboring gene despite our extensive efforts to identify such genes. We cannot yet exclude the possibility that these pCRMs have an in vivofunction related to their observed expression patterns. However, the poor conservation of these elements in D. pseudoobscurasuggest that they do not have a regulatory function, and raises the possibility that some \'random\' clusters of binding sites (that occur by chance or perhaps through selection on some functionally unrelated sequence feature) have the necessary characteristics to be active enhancers in the proper genomic environment (that is, near a promoter and not silenced by trans-acting chromatin mechanisms). That any such sequences exist suggests that the compositional and architectural constraints on binding sites in enhancers may be fairly weak. Whatever the nature of these constraints, it is clear that binding-site density is not the sole defining characteristic of functional enhancers. However, it is a surprisingly effective distinguishing one, and the usefulness of this and related methods \[[@B48]\] suggests that the broader application of such methods to different collections of transcription factors will be extremely valuable in annotating the regulatory content of animal genomes. New enhancers ------------- We identified double-stripe enhancers for ftzand odd. ftzand oddare generally classified as \'secondary\' pair-rule genes whose expression is governed by other pair-rule genes rather than by the maternal and gap transcription factors that govern the so-called \'primary\' pair-rule genes (eve, hand runt) (\[[@B49]\]; also reviewed in \[[@B50]\]). However, the ftzand oddenhancers described here were identified on the basis of binding sites for maternal and gap transcription factors, and function like the enhancers of primary pair-rule genes in directing expression in specific stripes. It has been suggested that the ftzenhancer is an evolutionary relic of the homeotic role played by ftzin primitive insects \[[@B51]\], a view supported by the apparently normal expression and activity of ftz when this element is missing. However, given our observation that non-functional binding sites clusters are not conserved, even over the relatively short evolutionary distance separating D. melanogasterand D. pseudoobscura, it seems unlikely that this element is purely vestigial. In fact, Yu and Pick \[[@B52]\] examined the expression pattern of the endogenous ftzgene and show that stripes 1 and 5 appear before other ftzstripes and they postulate the existence of stripe-specific regulatory elements that may exist outside of the characterized zebraand upstream elements such as the one identified and characterized in this study. The conservation of binding sites in both the ftzand oddenhancers suggest that they play an important role in development, and further call into question the distinction between primary and secondary pair-rule genes. Two of the new enhancers (CE8011 and CE8012) are adjacent to and apparently regulate two linked genes with very similar patterns of embryonic expression. Both nub(also known as pdm1) and pdm2are expressed in the anterior and posterior midgut primordium and in neuroblasts. CE8011, found immediately upstream of nub, regulates its early expression, and not its later neuroblast expression. In contrast, CE8012, found in an intron of pdm2regulates its expression only in neuroblasts and not earlier. While we did not detect a neuroblast enhancer for nubor a blastoderm enhancer for pdm2in our single-species binding-site cluster search, a number of interesting pdm2regions were discovered in our eCIS-ANALYST search (two are listed in Table [4](#T4){ref-type="table"}). Regulatory models and improving the accuracy of CRM prediction -------------------------------------------------------------- The accuracy of our enhancer predictions would almost certainly be improved if we restricted our search space to genomic regions adjacent to genes known to be regulated by particular transcription factors. Drosophilaenhancers have been known to work at distances of up to 100 kb, but most are within 10 kb of their target gene. All of our true-positive predictions were within 10 kb of the known or predicted transcription start site of a gene with a pattern that was known, or plausibly could have been, regulated by the five regulators used in our screen (anterior-posterior patterns in the blastoderm; expression in neuroblasts). In contrast, only one of the negative predictions was this close to such a gene - an additional four were within 50 kb. As the comprehensive atlas of embryonic expression patterns is completed \[[@B21],[@B53]\] it will be possible to restrict searches for CRMs to regions of the genome near genes with expression patterns that could arise from the regulators being considered, or to prioritize the results of whole-genome screens on the basis of whether they are near plausible targets. Comprehensive methods for inferring regulatory interactions where they are not already known will be critical for the widespread application of binding-site clustering methods. In addition to allowing less stringent focused screens, they will also help overcome the combinatorial challenge raised by the existence of up to 700 sequence-specific transcription factors in Drosophila. Even assuming the availability of binding data for all of these factors, it will not be possible to search for targets of all combinations of these factors - there are too many possibilities. This is not just a practical problem - it is a fundamental statistical problem. While the false-positive rate for a single combination of factors is low, if we tried even all pairs of factors, it is likely that every region of the genome would have a high binding-site density for some collection of factors. Sequence data from other Drosophilaspecies may allow us to determine which of these collections are conserved and therefore likely to be functional, but it is unlikely that all aspects of regulation can be inferred from comparative analyses and therefore it is essential that we continue to dissect the regulatory network by traditional means. A greater current limitation in the widespread application of binding-site clustering methods is the absence of high-quality binding data for most Drosophilatranscription factors. The initial success of methods that use in vitrobinding data to predict regulatory targets has prompted the characterization of binding specificities for many additional factors. However, the heterogeneity of approaches used makes it difficult to combine these data in an optimal manner. In addition, most of the available transcription factor binding data consists of a few to several dozen high-affinity sites. While these data are very useful, they do not fully represent the binding capacity of a factor and thus do not permit the identification of intermediate or low-affinity sites which are known to be important in some regulatory systems \[[@B54]\]. We have begun to apply high-throughput methods \[[@B55]\] to characterize a broad spectrum of target sites for all of the transcription factors involved in early embryogenesis. The results will ultimately allow us to estimate the binding affinity of each factor for any target sequence. Comparative genomics in CRM predictions --------------------------------------- The extent of non-coding sequence conservation between D. melanogasterand D. pseudoobscurawas surprising. A major motivation for the National Human Genome Research Institute (NHGRI) support of the D. pseudoobscuragenome sequencing was the identification of conserved regions that would guide the annotation of functional sequences in D. melanogaster. D. pseudoobscurawas chosen as the second member of this genus to be sequenced in part because it was felt that it had separated from D. melanogastersufficiently long ago that non-functional sequences would exhibit substantial divergence. However, despite an evolutionary separation that is greater than human and mouse (an average synonymous substitution rate of 1.8-2.6 substitutions/site \[[@B29]\] compared to 0.6 substitutions/site \[[@B30]\]), and despite some variation in conservation in non-coding sequences, we were not able to use standard measures of sequence conservation to differentiate active pCRMs from their flanking sequence or from inactive pCRMs, reinforcing other recent observations \[[@B32]\]. One reason for the limited efficacy of these methods is that they do not recognize the specific patterns of conservation characteristic of different classes of functional sequences. For example, coding sequences can be easily recognized from the characteristic triplet pattern in evolutionary rates where the third (and often synonymous) position of codons tends to evolve at a greater rate than the first two positions \[[@B56],[@B57]\]. Similarly, RNAs that form conserved secondary structures can be recognized by patterns of co-substitution (\[[@B58]\] and references cited within). The early developmental enhancers we are studying here are made up of large collections of transcription factor-binding sites, and it is expected that both individual functional binding sites and the overall composition of functional CRMs will be conserved \[[@B25],[@B26]\]. Conservation of binding-site clustering is a specific evolutionary signature of this class of functional regulatory sequences, and, like the evolutionary signatures of protein-coding and RNA genes, can be used to specifically identify these sequences from comparative sequence data. Contrast PCE8010 (the oddstripe enhancer) and PCE8015 (Figure [3](#F3){ref-type="fig"}). Both have the same overall amount of sequence conservation, indicating that they are under some functional constraint. However, 80% of the predicted binding sites in PCE8001 are conserved, compared to 20% for PCE8015. The conservation of binding sites (both number and location) in PCE8001 makes it highly unlikely that the cluster was found by chance in D. melanogaster, and suggests (correctly) that this sequence is actively responding to the presence of these binding sites. The poor conservation of binding sites in PCE8015 (no greater than is found in random regions of genome) suggests either that the BCD, HB, KR, KNI and CAD sites in this region are not functional or that the region is undergoing rapid functional diversification. Of course the absence of binding site conservation does not suggest that the sequence is non-functional, merely that these sequences are unlikely to have the particular function we are studying here. From the data shown in Figure [4](#F4){ref-type="fig"}, we expect the incorporation of binding-site conservation into the CRM search process to greatly reduce the number of false-positive predictions. We anticipate that a significant number of the new predictions from our genome-wide screen and screen targeted at genes with early anterior-posterior patterns to be active CRMs, and we have begun testing these predictions. The pattern of binding-site conservation in positive pCRMs sheds additional light on the processes that govern CRM evolution. We find that predicted binding sites in positive D. melanogasterpCRMs are roughly three times more likely to be aligned to predicted sites in the D. pseudoobscuracompared to predicted binding sites in negative pCRMs, in the sequences flanking pCRMs, or in random regions of the genome. The demonstration that this strictest form of binding-site conservation is strengthened in functional CRMs contrasts with an earlier study that concluded that binding sites in functional CRMs had only a slightly elevated probability of falling in conserved sequence \[[@B32]\]. Their methodology differed from ours in that they used randomly shuffled binding-site positions within functional CRMs as the background, while we used actual predicted binding-site positions in randomly picked regions of the genome. In addition to this colinear conservation, we also observe that there is an overall enrichment for binding sites in positive pCRMs independent of the conservation of individual sites. Specifically, the presence of a binding site for a factor in a positive D. melanogasterpCRM increases (relative to negative pCRMs and random genomic fragments) the probability of finding a site for the same factor in the orthologous region of D. pseudoobscura, even if the site is not in the same (aligned) position. Thus, in this set of positive pCRMs, there appears to be selection to maintain binding site composition, but not always the specific order and orientation of sites. This is consistent with models of enhancer plasticity that have been proposed and discussed elsewhere \[[@B25],[@B59]-[@B61]\]. The relative importance of binding-site architecture and binding-site composition to maintaining the function of an enhancer over evolutionary time remains unclear. Over relatively short evolutionary distances (as between D. melanogasterand D. pseudoobscura) most binding sites are conserved and found in the same place. Over longer evolutionary distances, individual binding sites are often poorly conserved even as the overall composition and function of a CRM is conserved. From a practical perspective, this requires adjusting how conservation is incorporated into searches for clusters of binding sites that are likely to be CRMs. For relatively short evolutionary distances, searches for clusters of aligned sites will be less sensitive to noise and will focus on functional binding sites. For longer distances, where binding site turnover will likely preclude searching for clusters of conserved sites, searches for conserved binding site clusters should still work well. In fact, this latter method can work - with some modification - among species whose sequences can no longer be aligned. Anopheles gambiaediverged from its common ancestor with D. melanogasterroughly 220 million years ago, and there is little or no detectable non-coding sequence similarity between these two species. Nonetheless, we find clusters of HB, KR and KNI binding sites in the vicinity of gap and pair-rule genes and suggest that many of these are functional orthologs of D. melanogasterCRMs. Despite strong selection to maintain function, enough binding-site turnover has occurred in these CRM during their 220 million years of independent evolution to eliminate detectable sequence similarity. But they remain functionally similar and we can detect this functional similarity through its evolutionary signature. With methods like the one we have presented here, aided by new and better binding data on Drosophilatranscription factors and an impending wealth of comparative sequence data, we anticipate rapid progress on the identification and functional characterization of regulatory sequences. We will then be able to turn our attention to the next great challenge - understanding the precise relationship between the binding-site composition and architecture of regulatory sequences and the expression patterns they specify. Materials and methods ===================== Collection of CRMs ------------------ The collection of CRM sequences was previously described \[[@B11]\] Transgenics ----------- DNA fragments identified as candidate CRMs were amplified from either bacterial artifical chromosome (BAC) or y; cn bw spfly genomic DNA by PCR using two primers containing unique sequence and synthetic AscI and NotI restriction sites (Additional data file 5). The PCR product was digested with AscI and NotI, and inserted in its native orientation into the AscI-NotI site of a modified CaSpeR-AUG-bgal transformation vector \[[@B62]\] containing the evebasal promoter, starting at -42 bp and continuing through codon 22 fused in-frame with lacZ\[[@B63]\]. The P-element transformation vectors were injected into w^1118^embryos, as described previously \[[@B63],[@B64]\]. Transgenic fly lines containing CRMs CE8005 (7A), CE8016 (55C) and CE8020 (70EF) were verified by generating genomic DNA \[[@B65]\] from each line for PCR. PCR products were amplified using primers designed from the CaSpeR-AUG-bgal vector - forward primer 5\' CGCTTGGAGCTTCGTCAC and reverse primer 5\' GAGTAACAACCCGTCGGATTC and 35 cycles (Gene Amp 9700, Perkin-Elmer). The resulting PCR products were sequenced using standard conditions with BigDye version 3.0 and electrophoresed on a 3730 capillary sequencer (ABI). Whole-mount in situhybridizations ----------------------------------- Embryonic whole-mount in situRNA hybridizations were performed as previously described \[[@B21]\]. RNA probes were generated using cDNA clones RE29225 (gt), RE14252 (odd), RE34782 (nub), RE49429 (pdm2), and RE47384 (sqz). Exon 1 of the ftzgene was amplified from genomic DNA using forward primer 5\' GCGTTGCGTGCACATC and reverse primer 5\' ATTCTTCAGCTTCTGCGTCTG. The PCR product was cloned into the TA vector (Invitrogen) and used to generate ftzRNA probe. Double-labeling --------------- RNA probes, using cDNAs or genomic DNA as templates, were labeled with fluorescein-12-UTP while lacZRNA probes were labeled with digoxigenin-11-UTP (Roche). Hybridizations were performed as described above with the following modifications: (1) 2 μl of each probe were added to give a final concentration of 1:50; (2) sequential alkaline phosphatase staining was performed first with Sigma Fast red to detect endogenous transcripts, stopped by washing for 30 min in 0.1 M glycine-HCl pH 2.2, 0.1% Tween-20 at room temperature, and then continued as described to detect lacZexpression. Assembly -------- The input to the genome assembly was the set of whole-genome shotgun reads from the Baylor Genome Sequencing Center retrieved from the National Center for Biotechnology Information (NCBI) Trace Archive, consisting of 2,607,525 total sequences. After trimming the sequences to remove vector and low-quality regions, the average read length was 607 bp. Approximately 75% of the reads were from short insert (approximately 2.5-3.0 kb) libraries, with another 25% from longer (6-7 kb) libraries. Another 46,040 reads came from the ends of 40-kb fosmids. We ran the Celera Assembler several times, and found that by adjusting one parameter in particular we could produce considerably better assemblies. In particular, the assembler has an arrival rate statistic j, which measures the probability that a contig is repetitive on the basis of its depth of coverage. The default setting is very conservative: if a contig has more than 50% likelihood of being repetitive, it is marked as such and is set aside during most of the assembly process. For large highly repetitive mammalian genomes this setting may be appropriate, but for D. pseudoobscurawe found that setting it to 90% or higher produced considerably better contigs, while apparently causing few if any misassemblies. The overall assembly contained 10,089 scaffolds and 10,329 contigs, containing 165,864,212 bp. The estimated span of the scaffolds, using the gap sizes estimated from clone insert sizes, is 172,362,884. The largest scaffold was 3.05 million base-pairs (Mbp) and the scaffold N50 size was 418,046. (The N50 size is the size of the smallest scaffold such that the total length of all scaffolds greater than this size is at least one half the total genome size, where genome size here is 172 Mbp.) There are 308 scaffolds larger than 100,000 bp, whose total span is 129.5 Mbp. The N50 contig size, using 166 Mbp as the genome size (not counting gaps), was 43,555. Another measure of assembly quality is the number of large contigs: if we define \'large\' as 10 kbp, then the assembly contains 3177 large contigs whose total length is 131,067,828 bp. (For reference, the assembly produced by the Baylor Human Genome Sequencing Center contains 129.4 Mbp in all contigs, including small ones, and the span of all scaffolds is 139.3 Mbp.) All of our contigs and scaffolds are freely available by anonymous ftp at \[[@B66]\]. Alignment and conservation of pCRMs ----------------------------------- The extent and pattern of conservation between D. melanogasterand D. pseudoobscurain regions containing pCRMs were determined as follows. The D. melanogastergenomic sequence of the region of interest (with known repetitive elements masked) was extracted from a BioPerl genome database \[[@B67]\] containing Release 3.1 sequence and annotations from the Berkeley Drosophila Genome Project \[[@B68]\]. Potentially orthologous D. pseudoobscuracontigs/scaffolds were identified using WU-BLAST 2.0 \[[@B69]\] using default parameters except for (-span1 -spsepqmax = 5000 -hspsepsmax = 5000 -gapsepmax = 5000 -gapsepsmax = 5000). High-scoring pairs (HSPs) with E-values less than 1e-20 were flagged as potential homologous regions. HSPs located more than 5,000 bp from each other in the D. melanogastersequence were treated as separate hits. After examining dot-plots of the hits, we noticed a large number of small, local inversions that were found in both our assembly and the assemblies released by the Baylor Human Genome Sequencing Center. We used BLASTZ \[[@B70]\]) to automatically identify inversions, and when necessary inverted the corresponding D. pseudoobscurasequence. Each D. pseudoobscurasequence was aligned to the D. melanogastercorresponding sequence using LAGAN 1.2 \[[@B43]\] with default settings. A total of 31 genomic loci of approximately 50 kb were examined; these regions contain 36 pCRMs (the eveand hloci contain three pCRMs each, and PCE8003 and PCE8004 are within 20 kb of each other). Twenty-eight regions had aligned D. pseudoobscurasequence that spanned all or most of the region. For three regions (PCE8002, PCE8003/8004 and PCE8009) we were not able to identify large regions of orthologous sequence; these were excluded from subsequent comparative analyses. Dot-plots of the alignments from all 30 regions are available at \[[@B42]\]. Scoring gross conservation of pCRMs ----------------------------------- The conservation of a specific genomic segment was scored as the fraction of D. melanogasterbases aligned to the identical base in aligned regions (percent identity). Scoring binding-site conservation of pCRMs ------------------------------------------ We used two definitions of binding-site conservation. A binding site was considered \'aligned\' if it overlaps a predicted D. pseudoobscurabinding site for the same factor in the LAGAN alignment. Only overlap, and not strict alignment, was required to compensate for small errors in the alignment. A non-aligned binding site was considered \'preserved\' if it could be matched to a D. pseudoobscurasite for the same factor within the bounds of the pCRM, allowing each D. pseudoobscurasite to be the match for only a single D. melanogastersite. The number of aligned plus preserved sites for each factor in a region is thus equal to the minimum number of sites for that factor in the two species. Generating an orthology map for genome searches ----------------------------------------------- To develop an orthology map for genome-wide searches, we used NUCmer \[[@B71]\] to align the Release 3 D. melanogastergenome (with annotated repetitive elements and transposable elements masked) and the D. pseudoobscurascaffolds described above. NUCmer was run with the command line parameters (-c 36 -g 10 \--mum -d 0.3 -l 9). NUCmer generated a collection of short, highly conserved regions of homology (\'anchors\') spaced on average every 1 kb throughout the D. melanogastergenome. Anchors flanking either side of a D. melanogasterregion of interest were used to pull out the corresponding D. pseudoobscuraregion, and additional flanking anchors were examined to ensure that the region was unambiguously orthologous. The region identified was re-aligned to the melanogaster region with LAGAN 1.2 using default settings. Random sampling of non-coding genome ------------------------------------ To characterize properties of non-coding sequences across the genome, we picked 4,000 1-kb segments of the D. melanogastergenome, sampled uniformly from all non-coding sequence. For 3,300 of these, we could find orthologous regions in D. pseudoobscura, and these were used to calculate the properties of random non-coding sequence shown in Figure [4](#F4){ref-type="fig"} and discussed in the text. Properties determined using this data are considered properties of only the portion of the genome that is detectably orthologous under our conditions. The regions themselves are available as supplemental material at \[[@B42]\]. eCIS-ANALYST genome searches ---------------------------- Binding-site clusters in the D. melanogastergenome were determined as described in \[[@B11]\], where the minimum number of sites (min\_sites) and the window size (wind\_size) are variable. Release 3 genomic sequence with exons masked was searched with PATSER \[[@B72]\] using the following command line options: -c -d2 -l4. An \'alphabet\' file (specified with the command line parameter \'-a\') was used to provide the following background frequencies: A/T = 0.297, G/C = 0.203. Position weight matrix (PWM) models were identical to those used in \[[@B11]\]. In the online version of eCIS-ANALYST, the minimum PWM match threshold site\_p is also variable, but in the current study it was held constant at 0.0003 for all factors. Tests using alternate values for this variable did not lead to significant improvement in prediction efficacy. For each potential D. melanogastercluster, we identified the corresponding D. pseudoobscuraregion using the homology anchors described above. A pairwise alignment was made using LAGAN 1.2 (default parameters), and the number of aligned and preserved binding sites were determined as described above. The 2-kb flanking either side of the pCRM was included in the alignment to avoid edge effects, and was subsequently removed when calculating pCRM properties. We examined our functional (positive) and non-functional (negative) pCRMs and noticed that in the positives, the lower bound for the number of conserved sites as a function of D. melanogastersites followed an approximately logarithmic curve (Additional data file 3). From this observation, we classified a D. melanogasterbinding site cluster as conserved if: ![](gb-2004-5-9-r61-i1.gif) where NS~m~is the number of binding sites in the D. melanogasterpCRM and NS~c~is the number of conserved binding sites. Different values of the logarithmic base bgive different behavior. The data shown in Additional data file 3 support values of bbetween 1.15 and 1.4. We defined a more intuitive parameter, CF(conservation factor), which can range from 0 to 1 where 0 is the least stringent threshold (b= 1.4) and 1 is the most stringent (b= 1.15) b= 1.4 - (CF\* (1.4 - 1.15)) (2) We performed genome searches with CFvalues of 0.25, 0.5, 0.55 and 0.75 and manually inspected the results with respect to false-negative and false-positive rates based on our 15 positive and 17 negative pCRMs (Additional data file 3). While we did not strictly optimize a single metric, we picked the values that gave a reasonable balance between false positives and false negatives, b= 0.25 for aligned sites alone, and b= 0.55 for aligned plus preserved sits. Genome-wide predictions ----------------------- eCIS-ANALYST genome searches were run with the following parameters: min\_sites = 10, wind\_size = 700 (run \#1), and min\_sites = 13, wind\_size = 1,100 (run \#2). All conserved clusters (with conservation defined as described in Equations 1 and 2 above) were combined. In order to capture weaker clusters, we performed an additional run (run number 3) using min\_sites = 9, wind\_size = 700. For this low stringency run, we used a non-standard conservation threshold different from the one described above, accepting all clusters with at least four aligned plus preserved sites, independent of the number of sites in D. melanogaster. We merged overlapping clusters from runs 1-3, yielding 929 non-overlapping clusters as described in Results. Four metrics were then used to rank these 929 pCRMs: the number of aligned binding sites; the density of aligned binding sites; the number of aligned plus preserved binding sites; and the density of aligned plus preserved binding sites. All values were normalized according to background distribution of random non-coding sequences. The four normalized values were then summed to compute an overall score, which was then renormalized to arrive at a final z-score used to rank pCRMs in Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"} and Additional data files 7, 8, 10, and 11. Additional data files ===================== The following additional data files are available with the online version of this article. Additional data file [1](#s1){ref-type="supplementary-material"} shows the binding site densities (column 1), aligned site densities (column 2), and aligned plus preserved site densities (column 3) for individual transcription factors. The top portion of each panel contains a histogram of the values for randomly chosen 1,000 bp regions of the D. melanogastergenome. The blue line plots the cumulative distribution. The colored asterisks show the average values for each class of pCRM. The panel below the histogram shows the values for each pCRM (each dot represents one pCRM, with positives in blue, negatives in red, ambiguous in green). Additional data file [2](#s2){ref-type="supplementary-material"} shows expression patterns of 65 genes adjacent to 122 pCRMs identified by eCIS-ANALYST. The images were obtained from the BDGP Embryonic Expression Pattern Database \[[@B33]\], and include all pCRMs from Additional data files 7,8,10,11 for which an adjacent gene had an early segmentation pattern. Additional data file [3](#s3){ref-type="supplementary-material"} shows discrimination of positive and negative pCRMs. Comparisons of the number of predicted binding sites in D. melanogasterpCRMs to the number of aligned sites (top panel) and aligned plus preserved sites (bottom panel). Blue dots represent the 15 positive pCRMs from the text; green dots the ten known CRMs that were below the threshold used in \[[@B11]\]; red dots negative pCRMs; pink dots ambiguous pCRMs. Gray boxes represent the distribution of values for random 1,000 bp non-coding regions. The blue line shows the discrimination function (see Materials and methods). Additional data file [4](#s4){ref-type="supplementary-material"} shows new pCRMs. Three 30 kb regions were chosen to illustrate new predictions: (A) the argos locus, (B) the CG4702 locus (note that CG31361 is not expressed in blastoderm embryos and PCE8494 is a low-scoring pCRM), and (C) the SoxN locus. Exons are shows as blue boxes, introns are represented with horizontal lines, and the direction of transcription is indicated by the arrow. New pCRMs are shown as gray ovals. The green graphs show average (in 300 bp windows) percent identity and fraction of bases in conserved blocks. Below the percent identity plots are shown insertions (gray boxes) and deletions (orange boxes) in the D. melanogastersequence relative to their D. pseudoobscuraortholog. The location of binding sites in D. melanogaster, binding sites in D. pseudoobscuraand aligned binding sites along with the density of sites averaged over 700 bp are shown in the bottom three panels for each region. Additional data file [5](#s5){ref-type="supplementary-material"} gives the primers used to amplify pCRMs for transgenics. Additional data file [6](#s6){ref-type="supplementary-material"} gives additional information from Table [2](#T2){ref-type="table"}. Additional data file [7](#s7){ref-type="supplementary-material"} gives all new pCRMs from genome-wide eCIS-ANALYST located within 20 kb of annotated transcript. Additional data file [8](#s8){ref-type="supplementary-material"} gives all new pCRMs from genome-wide eCIS-ANALYST located more than 20 kb from annotated transcript. Additional data file [9](#s9){ref-type="supplementary-material"} lists genes with anterior-posterior patterns and the source of the information. Additional data file [10](#s10){ref-type="supplementary-material"} gives all new pCRMs from genome-wide eCIS-ANALYST located within 20 kb of gene with anterior-posterior pattern. And, finally, Additional data file [11](#s11){ref-type="supplementary-material"} gives all new pCRMs from genome-wide eCIS-ANALYST located between 20 kb and 50 kb from gene with anterior-posterior pattern. Supplementary Material ====================== ::: {.caption} ###### Additional data file 1 The binding site densities (column 1), aligned site densities (column 2), and aligned plus preserved site densities (column 3) for individual transcription factors ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 2 Expression patterns of 65 genes adjacent to 122 pCRMs identified by eCIS-ANALYST ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 3 Discrimination of positive and negative pCRMs. Comparisons of the number of predicted binding sites in D. melanogasterpCRMs to the number of aligned sites (top panel) and aligned plus preserved sites (bottom panel) ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 4 New pCRMs ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 5 The primers used to amplify pCRMs for transgenics ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 6 Additional information from Table 2 ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 7 All new pCRMs from genome-wide eCIS-ANALYST located within 20 kb of annotated transcript ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 8 All new pCRMs from genome-wide eCIS-ANALYST located more than 20 kb from annotated transcript ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 9 Genes with anterior-posterior patterns and the source of the information ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 10 All new pCRMs from genome-wide eCIS-ANALYST located within 20 kb of gene with anterior-posterior pattern ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 11 All new pCRMs from genome-wide eCIS-ANALYST located between 20 kb and 50 kb from gene with anterior-posterior pattern ::: ::: {.caption} ###### Click here for additional data file ::: Acknowledgements ================ We thank Richard Weiszman, Naomi Win and Nipam Patel for assistance with RNA in situhybridizations, Pavel Tomancak for generating the database to store images of stained transgenic embryos and Amy Beaton and members of the Hartenstein lab for discussions of embryonic patterns of expression, Casey Bergman and Joseph Carlson for generating the database to store CRM transgenic sequences and the members of the BDGP for clones and sequencing support. We also thank Arthur Delcher and Mihai Pop for help with running and fine-tuning the Celera Assembler. This work was supported by National Institutes of Health Grants HG00750 (to G.M.R.), and HL667201 (to M.B.E.), and LM06845 (to S.L.S.); Department of Energy contract DE-AC03-76SF00098 (to M.B.E.); and by the Howard Hughes Medical Institute. M.B.E. is a Pew Scholar in the Biomedical Sciences. Author contributions are as follows: B.D.P. made P-element constructs containing the 28 candidate CRMs. T.R.L. injected these constructs into Drosophilaembryos, screened for transformants and generated the lines for analysis. B.D.P. collected embryos, generated probes and performed whole-mount in situhybridization. B.D.P. and S.E.C. imaged and analyzed transgenic embryos. S.L.S. assembled the D. pseudoobscuragenomic sequence. B.P.B. and M.B.E. performed all computational analyses. S.E.C., M.B.E. and G.M.R. provided guidance and direction for the project. S.E.C. supervised experimental aspects of the project. M.B.E. supervised computational aspects of the project. M.B.E. wrote the paper. B.P.B. prepared the tables and figures. B.D.P. and S.E.C. contributed to the content and edited the paper. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Expression patterns of active pCRMs. Embryonic whole-mount in situRNA hybridizations using lacZprobe of transgenes with positive expression in independent lines (see Materials and methods). The first column (wild type) shows the endogenous gene expression; the second column (lacZ) shows transgene expression patterns; the third column shows double-labeled embryos with the endogenous (red) and transgene (blue) expression patterns. To the right of the images are maps of the gene regions centered on each pCRM. ::: ![](gb-2004-5-9-r61-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Predicted and aligned binding sites in pCRMs. Predicted binding sites and aligned binding sites (see Materials and methods) in positive, ambiguous and negative pCRMs (the positions of overlapping sites were adjusted slightly so that all sites could be seen). ::: ![](gb-2004-5-9-r61-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Binding-site conservation, but not sequence conservation, correlates with pCRM activity. Three 25-kb regions were chosen to illustrate patterns of sequence conservation and binding-site conservation. (a)even-skipped(eve) contains five previously known segmentation enhancers (labeled eve3/7, eve2, eve4/6, eve1, and eve5); (b)odd-skipped(odd) contains a single functional (positive) pCRM (CE8010); and (c)pipsqueak(psq) contains a non-functional (negative) pCRM (CE8015). Annotated genes are shown in blue, and the direction of transcription is indicated by the arrow. Gray ovals indicate experimentally tested fragments, and shaded gray boxes show the extent of pCRMs as defined by CIS-ANALYST (minimum of 13 sites within a 700 bp window). The green graphs show average percent identity (in 100-bp windows). Below the percent identity plots are shown insertions (gray boxes) and deletions (orange boxes) of 80 or more bp in the D. melanogastersequence relative to their D. pseudoobscuraortholog. The location of binding sites in D. melanogaster, binding sites in D. pseudoobscuraand aligned binding sites along with the average density of sites (700-bp windows) are shown in the bottom three panels for each region. \* in (a) indicates a new prediction (PCE8100). ::: ![](gb-2004-5-9-r61-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Conservation of clustering distinguishes positive and negative pCRMs. Each panel compares positive, negative and ambiguous pCRMs and random 1,000-bp non-coding regions based on (a)binding site density in D. melanogaster, (b)percent identity, (c)density of aligned sites, and (d)density of aligned plus preserved sites. The top portion of each panel contains a histogram of the values for randomly chosen 1,000-bp regions of the D. melanogastergenome. The blue line plots the cumulative distribution. The colored asterisks show the average values for each class of pCRM. The unshaded panel below the histogram shows the values for each pCRM (each dot represents one pCRM, with positives in blue, negatives in red, ambiguous in green). The shaded panel at the bottom shows the average value for 1,000-bp non-coding sequences within 20 kb of each pCRM. ::: ![](gb-2004-5-9-r61-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Inclusion of evolutionary information greatly increases the specificity and selectivity of CRM searches based on binding-site clustering. The effects of integrating comparative data into searches for binding site clusters were assessed by counting the number of (a)true positive, (b)negative and (c)novel CRMs recovered at the different site density cutoffs plotted on the x-axis. The positives used here include the 15 positive pCRMs from Table 2 and 10 additional positive CRMs from the literature (see text), all of which have identifiably orthologous sequence in D. pseudoobscura, while the negatives included only the 14 non-functional pCRMs for which orthologous sequence in D. pseudoobscuracould be found. The solid line in each panel shows the results without the use of D. pseudoobscura; the dashed line shows the results with D. pseudoobscura. Searches displayed were performed using the aligned sites constraint (see Materials and methods). Comparable results were obtained for the aligned + preserved sites constraint. The number of false positives is not strictly monotonically decreasing with an increasing binding site cutoff. This stems from the cluster merging behavior of CIS-ANALYST - sometimes a decrease in the minimum number of sites leads CIS-ANALYST to tack on a lower-density cluster that is adjacent to a higher-density one, resulting in a single cluster with more sites but lower site density. This can actually increase the number of conserved sites necessary to reach the conservation threshold (see Materials and methods). ::: ![](gb-2004-5-9-r61-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Expression patterns of genes adjacent to high-scoring pCRMs. Wild-type embryonic expression patterns of 36 genes adjacent to 53 pCRMs identified by eCIS-ANALYST (see Tables 3 and 4). The images were obtained from the BDGP Embryonic Expression Pattern Database \[33\], and include all pCRMs from Tables 3 and 4 for which an adjacent gene had an early segmentation pattern. ::: ![](gb-2004-5-9-r61-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Genomic location of pCRMs and neighboring genes ::: pCRM ID\* Name CRM activity Arm pCRM start pCRM end pCRM length 5\' gene pCRM relative position 3\' gene pCRM relative position ---- --------- ------ --------------------- -------------- ----- ------------ ------------ ------------- ------------ ------------------------ ------------ ------------------------ 1 PCE7001 runt stripe 3 \+ X 20,357,206 20,358,294 1,089 CG1338 -9,550 run -8,561 2 PCE7002 eve stripes 3/7 \+ 2R 5,035,494 5,036,771 1,278 CG12134 3,713 eve -2,952 3 PCE7003 eve stripe 2 \+ 2R 5,038,454 5,039,040 587 CG12134 6,673 eve -683 4 PCE7004 eve stripes 4/6 \+ 2R 5,044,597 5,045,395 799 eve 4,874 TER94 -4,398 5 PCE7005 hairy stripe 7 \+ 3L 8,624,351 8,625,245 895 CG6486 16,118 h -9,423 6 PCE7006 hairy stripe 6 \+ 3L 8,625,452 8,626,319 868 CG6486 17,219 h -8,349 7 PCE7007 hairy stripes 1,5 \+ 3L 8,629,180 8,629,966 787 CG6486 20,947 h -4,702 8 PCE7008 kni upstream \+ 3L 20,615,070 20,616,425 1,356 kni -1,169 CG13253 21,311 9 PCE7009 hb HZ1.4 \+ 3R 4,526,315 4,527,521 1,207 hb -2,760 CG8112 403 10 PCE8001 1 gt posterior domain \+ X 2,187,439 2,188,382 944 gt -1,704 tko 12,366 11 PCE8010 2 odd stripes 3/6 \+ 2L 3,601,750 3,602,509 760 odd -2,433 Dot -9,351 12 PCE8011 3 nub blastoderm \+ 2L 12,605,345 12,606,039 695 CG15488 2,687 nub -1,178 13 PCE8024 4 ftz stripes 1/5 \+ 3R 2,693,713 2,694,405 693 ftz 3,667 Antp 131,873 14 PCE8012 5 pdm2 neurogenic \+ 2L 12,663,878 12,664,600 723 *pdm2* 2,875 *pdm2* 2,875 15 PCE8027 6 sqz neurogenic \+ 3R 15,000,096 15,000,905 810 sqz 10,137 CG14282 -1,833 16 PCE8005 7 cluster\_at\_7A amb. X 6,996,209 6,996,756 548 CG32725 -17,671 CG1958 -10,524 17 PCE8016 8 cluster\_at\_55C amb. 2R 13,354,407 13,355,109 703 CG14502 957 CG14502 957 18 PCE8020 9 cluster\_at\_70F amb. 3L 14,665,967 14,666,676 710 ome 10,334 ome 10,334 19 PCE8006 13 cluster\_at\_7B \- X 7,239,486 7,240,124 639 CG11368 46,902 CG32719 13,096 20 PCE8008 15 cluster\_at\_8F \- X 9,457,631 9,458,375 745 btd 24,460 Sp1 -33,567 21 PCE8013 17 cluster\_at\_34E \- 2L 13,989,283 13,990,132 850 rk -5,879 bgm -5,767 22 PCE8014 18 cluster\_at\_36F \- 2L 18,400,758 18,401,458 701 CG31749 36,362 RpS26 19,862 23 PCE8015 19 cluster\_at\_47A \- 2R 5,664,440 5,665,094 655 *psq* 45,904 *psq* 45,904 24 PCE8017 20 cluster\_at\_56B \- 2R 14,266,629 14,267,261 633 CG7097 24,156 CG7097 24,156 25 PCE8018 21 cluster\_at\_59B \- 2R 17,995,894 17,996,609 716 CG32835 759 CG32835 759 26 PCE8019 22 cluster\_at\_67B \- 3L 9,529,913 9,530,579 667 CG32048 10,499 CG32048 10,499 27 PCE8021 23 cluster\_at\_75C \- 3L 18,339,914 18,340,665 752 grim -86,621 rpr 6,617 28 PCE8022 24 cluster\_at\_76C \- 3L 19,594,180 19,594,883 704 CG8786 -1,409 CG8782 4,923 29 PCE8023 25 cluster\_at\_84A \- 3R 2,595,162 2,595,926 765 Ama 6,847 Dfd -21,632 30 PCE8025 26 cluster\_at\_85C \- 3R 4,944,607 4,945,444 838 pum 117,315 pum 117,315 31 PCE8026 27 cluster\_at\_88F \- 3R 11,424,315 11,424,996 682 CG18516 -45,803 CG5302 -33,626 32 PCE8028 28 cluster\_at\_95C \- 3R 19,757,908 19,758,531 624 Gdh 950 Gdh 950 33 PCE8003 11 cluster\_at\_5C.1 \- X 5,658,504 5,659,131 628 CG3726 952 CG3726 952 34 PCE8004 12 cluster\_at\_5C.2 \- X 5,674,913 5,675,606 694 CG3726 17,361 CG3726 17,361 35 PCE8009 16 cluster\_at\_12E \- X 14,146,556 14,147,218 663 CG32600 93,317 CG32600 93,317 36 PCE8002 10 cluster\_at\_4B \- X 4,124,119 4,125,459 1,341 CG12688 2,032 CG32773 3,408 37 PCE8007 14 cluster\_at\_7F Unknown X 8,350,658 8,351,315 658 Caf1-180 -5,486 oc 38,281 \IDs in this column are taken from \[11\]. Genomic locations of the 37 pCRMs identified in our previous genome search. All coordinates are from D. melanogasterRelease 3 \[68\]. pCRMs 1-9 were reported prior to our original search, and we attempted to characterize 10-37 in the current study (we reported PCE8001 in our previous publication). pCRMs10-15 recapitulate endogenous expression patterns of embryonic genes, and 16-18 drive ambiguous (amb.) expression patterns, as described in the text. pCRMs 19-36 drove no detectable expression in the embryo, and pCRM 37 was not tested. Orthologous regions were identified in D. pseudoobscurafor all but pCRMs 33-37. The 5\' and 3\' gene columns correspond to the closest transcription (or annotation) start 5\' and 3\' of the pCRM. If a pCRM is within an intron, only the intron-containing gene is reported and its name is given in italics. The names of genes with early anterior-posterior patterns are in bold. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Sequence and binding-site conservation in pCRMs between D. melanogasterand D. pseudoobscura* ::: pCRM Name CRM activity pCRM length (D. melanogaster) pCRM length (D. pseudoobscura) Percent identity D. melanogastersites D. pseudoobscurasites Conserved sites Fraction conserved ---- -------------------- --------------------- -------------- --------------------------------- ---------------------------------- ------------------ ------------------------ ------------------------- ----------------- -------------------- ----- ----- 1 PCE7001 runt stripe 3 \+ 1,089 1,504 71% 27 20 11 20 41% 74% 2 PCE7002 eve stripes 3/7 \+ 1,278 1,114 61% 28 25 21 25 75% 89% 3 PCE7003 eve stripe 2 \+ 587 771 67% 14 10 9 10 64% 71% 4 PCE7004 eve stripes 4/6 \+ 799 1,003 70% 20 18 13 17 65% 85% 5 PCE7005 hairy stripe 7 \+ 895 869 66% 20 16 12 16 60% 80% 6 PCE7006 hairy stripe 6 \+ 868 952 62% 23 19 11 19 48% 83% 7 PCE7007 hairy stripes 1,5 \+ 787 723 56% 16 15 9 13 56% 81% 8 PCE7008 kni upstream \+ 1,356 1,654 68% 33 31 24 30 73% 91% 9 PCE7009 hb HZ1.4 \+ 1,207 1,383 69% 24 23 17 21 71% 88% 10 PCE8001 gt posterior domain \+ 944 1,092 64% 23 19 15 18 65% 78% 11 PCE8010 odd stripes 3/6 \+ 760 825 70% 17 19 12 16 71% 94% 12 PCE8011 nub blastoderm \+ 695 894 70% 18 13 10 12 56% 67% 13 PCE8024 ftz stripes 1/5 \+ 693 744 77% 14 10 10 10 71% 71% 14 PCE8012 pdm2 neurogenic \+ 723 723 72% 14 8 4 8 29% 57% 15 PCE8027 sqz neurogenic \+ 810 818 69% 16 17 11 14 69% 88% 16 PCE8005 cluster\_at\_7A amb. 548 819 54% 13 4 2 2 15% 15% 17 PCE8016 cluster\_at\_55C amb. 703 1,617 55% 16 6 3 6 19% 38% 18 PCE8020 cluster\_at\_70F amb. 710 538 47% 14 2 2 2 14% 14% 19 PCE8006 cluster\_at\_7B \- 639 663 69% 15 9 8 8 53% 53% 20 PCE8008 cluster\_at\_8F \- 745 716 58% 14 2 1 2 7% 14% 21 PCE8013 cluster\_at\_34E \- 850 919 61% 17 8 6 8 35% 47% 22 PCE8014 cluster\_at\_36F \- 701 596 53% 15 6 5 6 33% 40% 23 PCE8015 cluster\_at\_47A \- 655 652 66% 16 3 3 3 19% 19% 24 PCE8017 cluster\_at\_56B \- 633 331 33% 15 9 4 8 27% 53% 25 PCE8018 cluster\_at\_59B \- 716 960 59% 16 4 3 4 19% 25% 26 PCE8019 cluster\_at\_67B \- 667 675 62% 15 7 5 6 33% 40% 27 PCE8021 cluster\_at\_75C \- 752 640 59% 19 13 10 12 53% 63% 28 PCE8022 cluster\_at\_76C \- 704 725 67% 15 9 7 9 47% 60% 29 PCE8023 cluster\_at\_84A \- 765 1,001 55% 16 7 5 7 31% 44% 30 PCE8025 cluster\_at\_85C \- 838 827 54% 16 6 1 5 6% 31% 31 PCE8026 cluster\_at\_88F \- 682 1,096 62% 16 6 5 5 31% 31% 32 PCE8028 cluster\_at\_95C \- 624 723 60% 15 6 4 6 27% 40% 33 PCE8003 cluster\_at\_5C.1 \- 628 None 15 34 PCE8004 cluster\_at\_5C.2 \- 694 None 15 35 PCE8009 cluster\_at\_12E \- 663 None 15 36 PCE8002 cluster\_at\_4B \- 1,341 None 28 37 PCE8007 cluster\_at\_7F Unknown 658 None 15 Mean (pCRMs 1-15) 899 1,005 67% 20 18 13 17 61% 80% Mean (pCRMs 19-32) 712 752 58% 16 7 5 6 30% 40% Conservation properties are listed for the pCRMs described in Table 1. The number and fraction of conserved sites are shown under two conditions - aligned sites only (A), or aligned + preserved sites (A+P) (see Materials and methods). D. pseudoobscurasequences used to determine these properties are available as supplemental material at \[42\]. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### New pCRMs from genome-wide eCIS-ANALYST (75 highest scoring predictions) ::: CRM Known element overlap Arm pCRM start pCRM end pCRM length 5\' gene pCRM relative position 3\' gene pCRM relative position Conserved sites Conserved site density z score Additional gap/pair-rule gene within 20 kb pCRM relative position ---- --------- -------------------------- ----- ------------ ------------ ------------- --------------- ------------------------ --------------- ------------------------ ----------------- ------------------------ --------- -------------------------------------------- ------------------------ ------------- -------- 1 PCE8050 h stripes 3/4,6,7 \[73\] 3L 8,622,879 8,626,839 3,961 CG6486 +14646 h -7829 36 62 9 16 20.1 2 PCE8051 kni upstream \[74\] 3L 20,614,714 20,617,020 2,307 kni -813 CG13253 +20716 25 31 11 13 13.2 3 PCE8052 nub blastoderm 2L 12,604,311 12,606,913 2,603 CG15488 +1653 nub -304 20 33 8 13 11.6 4 PCE8053 eve stripes 3/7 \[75\] 2R 5,035,493 5,037,290 1,798 CG12134 +3712 eve -2433 21 24 12 13 11.5 Adam +5901 5 PCE8054 hairy stripes 1,5 \[73\] 3L 8,628,846 8,631,011 2,166 CG6486 +20613 h -3657 17 29 8 13 10.5 6 PCE8055 runt stripe 3 \[76\] X 20,356,848 20,360,054 3,207 CG1338 -9192 run -6801 17 34 5 11 10.3 7 PCE8056 X 20,323,964 20,326,397 2,434 CG11692 -12536 Cyp6v1 -4186 16 28 7 12 9.6 8 PCE8057 hb HZ1.4 \[77\] 3R 4,526,225 4,527,991 1,767 hb -2670 CG8112 +1273 17 21 10 12 9.5 9 PCE8059 eve stripes 4/6 \[78\] 2R 5,044,597 5,046,030 1,434 eve +4874 TER94 -3763 15 18 10 13 9.0 Adam +15005 10 PCE8060 gt posterior \[11\] X 2,186,709 2,189,069 2,361 gt -974 tko +11679 18 21 8 9 8.9 11 PCE8061 X 3,169,806 3,172,348 2,543 CG12535 -17954 CG14269 +21857 13 29 5 11 8.8 12 PCE8063 CE8021 3L 18,339,914 18,341,941 2,028 grim -86621 rpr +5341 16 20 8 10 8.5 13 PCE8064 3R 6,255,663 6,256,945 1,283 CG6345 -13879 Cyp12e1 -3594 13 17 10 13 8.4 14 PCE8065 3R 4,026,032 4,027,816 1,785 grn -18853 CG7800 -15898 15 19 8 11 8.4 15 PCE8066 X 20,348,460 20,352,624 4,165 CG1338 -804 run -14231 16 28 4 7 8.3 16 PCE8067 ftz upstream \[23\] 3R 2,682,314 2,684,591 2,278 Scr -7972 ftz -5455 15 22 7 10 8.3 17 PCE8068 X 18,701,007 18,702,700 1,694 CG32541 +39691 CG32541 +39691 12 22 7 13 8.2 18 PCE8069 2R 17,274,311 17,276,017 1,707 CG3380 -2521 dve -11496 14 19 8 11 8.2 19 PCE8070 2L 7,616,050 7,618,366 2,317 CG6739 +15430 CG13792 +19862 14 23 6 10 8.1 20 PCE8071 sqz neurogenic 3R 14,999,463 15,001,552 2,090 sqz +9504 CG14282 -1186 12 24 6 11 8.0 nos +16485 21 PCE8072 X 5,674,422 5,676,386 1,965 CG3726 +16870 CG12728 -6597 11 24 6 12 7.8 22 PCE8073 2R 14,903,099 14,903,925 827 Toll-7 +12482 Obp56i -27903 11 11 13 13 7.8 23 PCE8074 3R 23,192,304 23,192,750 447 CG13980 +8073 side +40862 7 8 16 18 7.7 24 PCE8075 3R 10,762,920 10,764,750 1,831 CG3837 +18501 CG14861 -75759 13 19 7 10 7.6 25 PCE8076 eve stripe 2 \[75\] 2R 5,038,454 5,039,041 588 CG12134 +6673 eve -682 8 10 14 17 7.6 Adam +8862 26 PCE8077 2L 13,541,662 13,542,651 990 kuz +9371 kuz +9371 11 13 11 13 7.6 27 PCE8078 2L 14,424,056 14,425,158 1,103 BG:DS06238.4 -16773 BG:DS08340.1 +7810 12 13 11 12 7.6 28 PCE8080 odd stripes 3/6 2L 3,601,045 3,602,748 1,704 odd -1728 Dot -9112 12 19 7 11 7.5 29 PCE8081 3L 17,412,324 17,413,414 1,091 CG18265 +24035 CG7603 -1413 11 14 10 13 7.5 30 PCE8083 3L 14,121,556 14,123,127 1,572 Sox21b -41352 D +4373 12 17 8 11 7.3 31 PCE8084 2L 4,098,489 4,099,006 518 ed +74542 ed +74542 7 9 14 17 7.3 32 PCE8085 2R 12,253,766 12,255,302 1,537 CG10953 -23540 CG10950 -3625 13 15 8 10 7.2 33 PCE8086 3L 20,612,647 20,614,073 1,427 *kni* +1254 CG13253 +23663 11 17 8 12 7.2 34 PCE8087 2R 3,391,037 3,391,561 525 CG30358 +10444 CG14755 -16724 7 9 13 17 7.2 35 PCE8088 3L 16,418,107 16,418,469 363 CG33158 +49435 argos +14111 6 6 17 17 7.2 36 PCE8089 3R 12,368,159 12,368,687 529 CG11769 +28970 CG31448 -670 7 9 13 17 7.2 CG14889 -13735 37 PCE8091 3L 11,213,064 11,213,664 601 scylla +3224 CG32083 +24695 8 9 13 15 7.1 38 PCE8092 2L 1,233,357 1,235,228 1,872 CG5156 +3715 CG5397 -6475 9 23 5 12 7.1 39 PCE8093 3L 15,688,222 15,691,204 2,983 comm -10920 CG13445 -67172 13 22 4 7 7.0 40 PCE8094 2R 10,492,861 10,493,546 686 CG30472 -5321 CG12959 -26488 9 9 13 13 7.0 41 PCE8095 3R 23,894,562 23,895,459 898 CG12870 +31901 CG12870 +31901 10 11 11 12 7.0 42 PCE8096 3L 6,762,543 6,765,157 2,615 vvl +12855 Prat2 +108336 13 20 5 8 6.9 43 PCE8097 3R 10,238,130 10,238,652 523 CG14846 -1983 CG14847 +4557 7 8 13 15 6.8 44 PCE8099 2L 18,305,051 18,306,251 1,201 Fas3 +6868 Fas3 +6868 10 14 8 12 6.7 45 PCE8100 eve early APR \[79\] 2R 5,042,174 5,042,884 711 eve +2451 TER94 -6909 8 10 11 14 6.7 Adam +12582 46 PCE8102 tll posterior \[80\] 3R 26,663,942 26,665,204 1,263 CG15544 +21005 tll -2251 11 13 9 10 6.6 47 PCE8104 ems neurogenic \[81\] 3R 9,723,602 9,724,936 1,335 E5 -23682 ems -2663 12 12 9 9 6.6 48 PCE8105 3R 17,817,909 17,818,791 883 Eip93F +25598 Eip93F +25598 9 11 10 12 6.6 49 PCE8106 3L 10,499,018 10,501,551 2,534 CG32062 +25485 CG32062 +25485 11 21 4 8 6.6 50 PCE8107 3L 4,612,891 4,614,005 1,115 CG13716 -161 CG13715 +1681 11 11 10 10 6.6 51 PCE8108 2L 14,403,771 14,404,937 1,167 CG15284 -4301 BG:DS06238.4 +2346 10 13 9 11 6.5 52 PCE8109 3R 7,941,601 7,942,426 826 CG31361 +17775 CG4702 +11512 9 10 11 12 6.5 53 PCE8110 2L 8,804,166 8,805,336 1,171 CG9468 -30684 SoxN -12519 10 13 9 11 6.5 54 PCE8111 3L 8,612,337 8,613,016 680 CG6486 +4104 h -21652 8 9 12 13 6.5 55 PCE8112 3L 4,377,989 4,379,208 1,220 CG7447 +13842 Syx17 -3984 11 12 9 10 6.5 56 PCE8113 2L 14,113,291 14,113,893 603 CG15292 -3974 CG31768 -6693 7 9 12 15 6.5 57 PCE8114 3L 3,997,600 3,998,923 1,324 CG14985 +13500 fd64A -799 11 13 8 10 6.5 58 PCE8115 eve stripe 1 \[79\] 2R 5,046,559 5,047,297 739 eve +6836 TER94 -2496 8 10 11 14 6.5 Adam +16967 59 PCE8116 2R 16,921,501 16,922,240 740 CG13493 -11091 PpN58A +4194 8 10 11 14 6.5 60 PCE8118 3R 14,822,848 14,823,484 637 *gukh* +13085 *gukh* +13085 8 8 13 13 6.4 61 PCE8119 3R 12,671,525 12,672,987 1,463 abd-A -15737 CG10349 -32477 11 14 8 10 6.4 62 PCE8120 3L 10,492,688 10,495,539 2,852 CG32062 +19155 CG32062 +19155 10 23 4 8 6.4 63 PCE8121 2L 16,841,696 16,842,392 697 CG6012 -2193 CG31781 -5178 8 9 11 13 6.4 64 PCE8122 3L 6,885,832 6,887,436 1,605 Prat2 -11445 CG14820 -5022 11 15 7 9 6.4 65 PCE8123 2L 15,162,778 15,164,524 1,747 BG:DS03192.2 -6373 BG:DS07295.1 +59479 11 16 6 9 6.4 66 PCE8124 2R 6,888,483 6,889,700 1,218 CG12443 +13963 CG13192 -428 10 13 8 11 6.4 67 PCE8125 2L 20,466,022 20,467,708 1,687 CG2493 -32831 CG15476 +4184 10 17 6 10 6.4 68 PCE8126 3L 2,779,198 2,779,658 461 CG2083 +1101 CG2083 +1101 6 7 13 15 6.3 69 PCE8127 X 4,630,473 4,632,106 1,634 CG12681 +14179 CG15470 -3196 9 18 6 11 6.3 70 PCE8128 3R 27,713,381 27,715,087 1,707 heph +35171 heph +35171 10 17 6 10 6.3 71 PCE8130 3R 12,383,752 12,385,269 1,518 *CG14889* +1858 *CG14889* +1858 11 14 7 9 6.3 72 PCE8131 3R 21,329,716 21,331,058 1,343 CG5111 +8355 msi -2351 8 17 6 13 6.3 73 PCE8132 3R 16,242,660 16,243,128 469 CG10881 +8657 CG17208 +20535 6 7 13 15 6.3 74 PCE8133 3R 24,120,296 24,122,240 1,945 CG12516 -668 larp +19112 12 15 6 8 6.2 75 PCE8134 3L 8,733,754 8,734,394 641 CG32030 +8601 CG32030 +8601 7 9 11 14 6.2 Seventy-five top pCRMs, ranked by a z-score based on the number and density of conserved binding sites (see text for details). Site density columns list the number of conserved sites per kilobase (relative to the D. melanogastersequence). The number and density of conserved sites are shown under two conditions - aligned sites only (A), or aligned + preserved sites (A+P) (see Materials and methods). The 5\' and 3\' gene columns correspond to the closest transcription (or annotation) start 5\' and 3\' of the pCRM. If a pCRM is within an intron, only the intron-containing gene is reported and its name is italicized. The names of genes with early anterior-posterior patterns are in bold. Early anterior-posterior genes that start within 20 kb of the pCRM (but are not the immediate annotation in the 5\' or 3\' direction) are also listed. Named enhancers without a reference are from this study. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Additional new pCRMs within 20 kb of genes with anterior-posterior patterns ::: CRM Known element overlap Arm pCRM start pCRM end pCRM length 5\' gene pCRM relative position 3\' gene pCRM relative position Conserved sites Conserved site density z score Additional Gap/pair-rule gene within 20 kb pCRM relative position ---- --------- ------------------------ ----- ------------ ------------ ------------- ---------------------------- ------------------------ ------------------------- ------------------------ ----------------- ------------------------ --------- -------------------------------------------- ------------------------ ------------ -------- 1 PCE8137 3R 12,053,627 12,055,472 1,846 *tara* +2239 *tara* +2239 10 17 5 9 6.1 2 PCE8139 2R 6,573,169 6,574,383 1,215 inv +32752 CG30034 +12378 10 12 8 10 6.1 en +19407 3 PCE8140 2R 15,167,055 15,168,270 1,216 CG16898 -98356 18w -6952 10 12 8 10 6.1 4 PCE8144 3L 3,503,831 3,504,156 326 Eip63E +7518 Eip63E +7518 4 6 12 18 6.1 ImpE2 -10525 5 PCE8145 3R 4,536,237 4,536,936 700 CG8112 +1795 CG8112 +1795 8 8 11 11 6.0 hb -12682 6 PCE8150 3R 6,379,567 6,380,474 908 *hth* +50936 *hth* +50936 8 11 9 12 6.0 7 PCE8165 X 8,390,109 8,392,075 1,967 oc -513 CG12772 -23984 10 16 5 8 5.8 8 PCE8166 3R 12,570,467 12,571,123 657 Ubx -10101 CG31275 +5951 7 8 11 12 5.7 9 PCE8167 Ubx S1 \[82\] 3R 12,589,099 12,589,755 657 CG31275 (Ubx adjacent) -11970 Glut3 -24295 7 8 11 12 5.7 10 PCE8169 ftz stripes 1/5 \[51\] 3R 2,693,336 2,694,915 1,580 ftz +3290 Antp +63624 11 12 7 8 5.7 11 PCE8170 3R 2,670,658 2,672,242 1,585 *Scr* +2100 *Scr* +2100 9 15 6 9 5.7 ftz -19388 12 PCE8177 2R 5,634,520 5,635,604 1,085 *psq* +4661 *psq* +4661 8 12 7 11 5.7 13 PCE8183 2L 7,305,525 7,305,940 416 *wg* +4205 *wg* +4205 5 6 12 14 5.6 14 PCE8187 2L 8,286,022 8,287,399 1,378 *Btk29A* +5904 *Btk29A* +5904 9 13 7 9 5.6 15 PCE8190 3L 6,589,453 6,590,721 1,269 *Glu-RI* +5891 *Glu-RI* +5891 9 12 7 9 5.6 16 PCE8193 Kr CD2 \[83\] 2R 20,268,656 20,269,940 1,285 CG9380 -36249 Kr -244 7 15 5 12 5.5 17 PCE8195 3L 5,126,445 5,126,805 361 *CG32423* +17297 *CG32423* +17297 4 6 11 17 5.5 18 PCE8198 2L 3,767,311 3,769,396 2,086 *bowl* +2110 *bowl* +2110 9 17 4 8 5.5 19 PCE8210 3L 7,925,371 7,926,049 679 exex +17651 RNaseX25 -4074 6 9 9 13 5.4 20 PCE8214 2L 12,601,146 12,602,225 1,080 ref2 -895 CG15488 -433 8 11 7 10 5.4 nub -6071 21 PCE8218 2L 10,545,226 10,547,197 1,972 *CG31721* +7937 *CG31721* +7937 10 14 5 7 5.3 22 PCE8226 2L 12,541,433 12,542,145 713 bun -11992 CG15489 -40512 6 9 8 13 5.2 23 PCE8235 X 2,190,216 2,191,697 1,482 gt -4481 tko +9051 9 12 6 8 5.2 24 PCE8237 2L 12,670,755 12,671,417 663 *pdm2* +3280 *pdm2* +3280 6 8 9 12 5.2 25 PCE8258 3L 15,491,385 15,492,925 1,541 *CrebA* +7093 *CrebA* +7093 7 15 5 10 5.1 26 PCE8270 3L 16,421,730 16,422,846 1,117 *argos* +9734 *argos* +9734 8 10 7 9 5.0 27 PCE8275 3L 18,329,419 18,330,261 843 grim -76126 rpr +17021 6 10 7 12 5.0 28 PCE8277 3R 6,448,750 6,449,993 1,244 *hth* +8759 *hth* +8759 6 14 5 11 5.0 29 PCE8297 2R 20,280,374 20,281,018 645 Kr +10190 CG30429 -9080 6 7 9 11 4.9 30 PCE8306 3L 12,278,550 12,279,346 797 CG4328 -28041 CG32105 -7436 6 9 8 11 4.9 31 PCE8307 3L 5,580,997 5,581,649 653 CG12756 -13449 CG5249 -8641 6 7 9 11 4.9 32 PCE8309 2L 3,825,809 3,827,419 1,611 slp1 +7561 slp2 -1991 8 13 5 8 4.9 33 PCE8314 2L 3,842,537 3,843,621 1,085 slp2 +13127 CG3964 -11628 6 12 6 11 4.8 34 PCE8328 2L 16,418,533 16,419,580 1,048 *BG:DS02780.1* +8016 Idgf1 -3783 7 10 7 10 4.8 35 PCE8331 3L 5,582,709 5,583,340 632 CG12756 -15161 CG5249 -6950 5 8 8 13 4.8 36 PCE8332 3R 2,725,376 2,726,195 820 *Antp* +32344 *Antp* +32344 6 9 7 11 4.8 37 PCE8338 3R 3,987,824 3,989,532 1,709 *grn* +17647 *grn* +17647 8 13 5 8 4.7 38 PCE8348 3L 18,966,181 18,967,380 1,200 *nkd* +26830 *nkd* +26830 7 11 6 9 4.7 39 PCE8355 3R 6,421,647 6,422,583 937 *hth* +8827 *hth* +8827 6 10 6 11 4.7 40 PCE8356 3L 22,244,275 22,244,894 620 *Ten-m* +80890 CG32450 -2161 6 6 10 10 4.7 41 PCE8358 3R 26,740,914 26,742,495 1,582 *Ptx1* +2496 *Ptx1* +2496 8 12 5 8 4.7 42 PCE8361 Ubx BRE \[84\] 3R 12,526,665 12,527,949 1,285 *Ubx* +32417 *Ubx* +32417 6 13 5 10 4.6 43 PCE8367 2R 4,771,288 4,771,881 594 CG10459 +3018 dap -1074 5 7 8 12 4.6 44 PCE8369 3L 14,540,753 14,541,382 630 *HGTX* +7066 *HGTX* +7066 6 6 10 10 4.6 45 PCE8370 3L 2,395,158 2,396,393 1,236 CG13800 +12412 CG32306 -13538 5 14 4 11 4.6 46 PCE8391 3L 5,254,002 5,254,895 894 CG32423 -16750 lama +55892 6 9 7 10 4.5 47 PCE8394 Kr 730 \[83\] 2R 20,266,323 20,267,047 725 CG9380 -33916 Kr -3137 6 7 8 10 4.5 48 PCE8398 3R 2,770,846 2,771,901 1,056 *Antp* +12307 *Antp* +12307 7 9 7 9 4.5 49 PCE8401 2L 12,660,502 12,661,614 1,113 CG15485 -2463 *pdm2* +5861 6 11 5 10 4.5 50 PCE8408 X 8,379,690 8,381,014 1,325 *oc* +8582 *oc* +8582 5 14 4 11 4.4 51 PCE8415 3R 13,867,601 13,868,164 564 CG7794 +18158 htl +6934 5 6 9 11 4.4 52 PCE8417 2L 587,804 588,638 835 *Gsc* +7714 *Gsc* +7714 6 8 7 10 4.4 53 PCE8418 3R 18,950,000 18,950,634 635 CG31457 -5638 *hh* +7739 5 7 8 11 4.4 cenB1A 12397 54 PCE8425 2R 18,693,096 18,694,318 1,223 *retn* +16917 CG5411 -6825 7 10 6 8 4.4 55 PCE8439 X 4,770,587 4,771,859 1,273 CG12680 +32240 ovo -17051 7 10 5 8 4.3 56 PCE8444 3L 18,330,763 18,332,045 1,283 grim -77470 rpr +15237 7 10 5 8 4.3 57 PCE8450 3L 5,141,131 5,141,793 663 *CG32423* +2971 CG10677 -438 5 7 8 11 4.3 58 PCE8458 3L 19,101,833 19,102,666 834 *fz2* +6194 [*fz2]{.underline} +6194 5 9 6 11 4.2 59 PCE8464 3L 17,314,105 17,314,815 711 tap* +5577 Cad74A +13577 6 6 8 8 4.2 60 PCE8483 2L 8,265,854 8,267,283 1,430 *Btk29A* +2646 *Btk29A* +2646 4 15 3 10 4.1 61 PCE8493 3R 6,403,852 6,405,604 1,753 *hth* +25806 *hth* +25806 7 12 4 7 4.1 62 PCE8494 3R 7,931,641 7,932,680 1,040 CG31361 +7815 *CG31361* +7815 6 9 6 9 4.1 63 PCE8495 2L 5,214,677 5,215,845 1,169 CG6514 +3847 *tkv* +14084 6 10 5 9 4.1 64 PCE8501 2L 5,247,719 5,248,767 1,049 *tkv* +10898 Cyp4ac1 -7804 6 9 6 9 4.1 65 PCE8511 3R 6,469,170 6,470,599 1,430 hth -4766 CG6465 +32311 7 10 5 7 4.0 66 PCE8512 pdm2 neurogenic 2L 12,663,453 12,664,721 1,269 *pdm2* +2754 pdm2 +2754 5 12 4 9 4.0 67 PCE8513 3L 14,550,945 14,551,746 802 HGTX -2497 Cyp314a1 -16963 5 8 6 10 4.0 68 PCE8515 2L 16,390,610 16,392,235 1,626 *BG:DS02780.1* +34314 *BG:DS02780.1* +34314 7 11 4 7 4.0 69 PCE8519 3L 8,975,309 8,975,873 565 *Doc2* +2077 *Doc2* +2077 5 5 9 9 4.0 Doc3 11402 70 PCE8520 2L 12,080,772 12,081,448 677 prd -5445 CG5325 -1193 4 8 6 12 4.0 71 PCE8521 2L 7,252,370 7,253,008 639 CG31909 +2569 Wnt4 +16391 5 6 8 9 4.0 Ndae1 -19639 72 PCE8528 X 14,366,706 14,367,311 606 *NetA* +17535 *NetA* +17535 4 7 7 12 4.0 73 PCE8531 3R 6,363,866 6,364,968 1,103 CG31394 -8970 *hth* +66442 6 9 5 8 4.0 74 PCE8533 3R 24,402,963 24,403,946 984 fkh -2792 Noa36 +10421 6 8 6 8 3.9 75 PCE8536 3R 12,764,472 12,765,970 1,499 *Abd-B* +4036 *Abd-B* +4036 7 10 5 7 3.9 Seventy-five top pCRMs within 20 kb of a gene with early anterior-posterior expression, excluding those already listed in Table 3, are ranked by a z-score based on the number and density of conserved binding sites (see text for details). Site density columns list the number of conserved sites per kilobase (relative to the D. melanogastersequence). The number and density of conserved sites are shown under two conditions - aligned sites only (A), or aligned + preserved sites (A+P) (see Materials and methods). The 5\' and 3\' gene columns correspond to the closest transcription (or annotation) start 5\' and 3\' of the pCRM. If a pCRM is within an intron, only the intron-containing gene is reported and its name is italicized. The names of genes with early anterior-posterior patterns are in bold. Early anterior-posterior genes that start within 20 kb of the pCRM (but are not the immediate annotation in the 5\' or 3\' direction) are also listed. Named enhancers without a reference are from this study. :::	PubMed Central	2024-06-05T03:55:48.467744	2004-8-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522868/", "journal": "Genome Biol. 2004 Aug 20; 5(9):R61", "authors": [ { "first": "Benjamin P", "last": "Berman" }, { "first": "Barret D", "last": "Pfeiffer" }, { "first": "Todd R", "last": "Laverty" }, { "first": "Steven L", "last": "Salzberg" }, { "first": "Gerald M", "last": "Rubin" }, { "first": "Michael B", "last": "Eisen" }, { "first": "Susan E", "last": "Celniker" } ] }
PMC522869	Background ========== Global gene-expression patterns are established and maintained by the concerted actions of transcription factors and the proteins that constitute chromatin. The global network of interactions between transcription factors and promoters in yeast is increasingly being characterized \[[@B1]\]. The role of chromatin in gene regulation is less clear, however. For example, the distribution of nucleosomes, the fundamental units of chromatin, is poorly understood on a gene-specific basis, much less a global basis \[[@B2]\]. The nucleosome consists of approximately 146 base-pairs (bp) of DNA wrapped around an octamer of histone proteins - two each of histones H2A, H2B, H3 and H4. Eukaryotic genomes are packaged into repeating units of nucleosomes separated by around 10-80 bp of linker DNA. High occupancy by nucleosomes is thought to be generally repressive \[[@B3]\], and extensive remodeling (and loss) of nucleosomes occurs in the promoters of genes undergoing activation \[[@B4]\]. In the case of the PHO5promoter in yeast, this remodeling proceeds until essentially no nucleosomes are detected across a region of several hundred base-pairs \[[@B5],[@B6]\]. Transcription factors and chromatin proteins each form complex regulatory networks that interact in a variety of ways \[[@B1],[@B7]\]. Transcription factors modify chromatin structure by recruiting enzymes that remodel nucleosomes or posttranslationally modify histones (by acetylation or methylation, for example) \[[@B8]-[@B10]\]. The modifications can be maintained through cell division and propagated to proximal nucleosomes by positive-feedback mechanisms \[[@B7],[@B11],[@B12]\]. Hence, a signal such as the activation of a transcription factor can be temporally and spatially transmitted through chromatin. Conversely, chromatin can influence transcription factor function by modulating the accessibility of target binding sites in the DNA \[[@B13],[@B14]\]. We used chromatin immunoprecipitation (ChIP) and DNA microarrays to evaluate nucleosome occupancy levels for essentially all promoters in yeast. Promoters that regulate active genes, contain multiple conserved motifs or recruit Rap1 tend to be relatively nucleosome-depleted. We also used real-time PCR and micrococcal nuclease digestion to show that nucleosomes are depleted in the vicinity of Rap1 consensus sites. This depletion can be partially reversed by the actions of the small molecule rapamycin or by removing Rap1-binding sites. We suggest that other transcription factors have less robust nucleosome-depleting activities than Rap1 and must therefore act collaboratively to gain access to their cognate sites in the DNA. Results ======= ChIP-based assay for nucleosome occupancy ----------------------------------------- Histones are essential components of the nucleosome and efficiently cross-link to nucleosomal DNA. Antibodies against invariant portions of histones have been used previously in ChIP assays to follow nucleosome loss at the yeast PHO5promoter \[[@B5],[@B6]\]. We extended this approach to evaluate relative nucleosome occupancy at essentially all promoters and other intergenic regions in yeast. DNA associated in vivowith histone H3 was isolated by ChIP using antibody against the carboxy terminus of histone H3 (no posttranslational modifications are thought to occur in this region). ChIP DNA and unenriched control DNA were amplified by in vitrotranscription and evaluated using microarrays. DNA associated with histone H2B was evaluated in a similar fashion using anti-FLAG antibody and a FLAG-H2B strain. H3 and H2B datasets were compiled by averaging four and three independent biological experiments, respectively. These datasets are remarkably similar as shown by a genome-wide correlation of 0.83 (Figure [1a-c](#F1){ref-type="fig"}). This correlation is comparable to that observed when comparing replicate H3 datasets (or H2B datasets), and suggests that both assays measure similar phenomena. In the H3 and H2B datasets, respectively, there are 347 and 214 regions depleted at least 1.5-fold relative to the average over all intergenics. In contrast, there are just 84 and 6 regions in the respective datasets enriched at least 1.5-fold relative to this average. The relatively narrow range of ChIP enrichment and the negative skew of the data (Figure [1b](#F1){ref-type="fig"}) are consistent with the conventional view that the majority of the genome is packaged into nucleosomes with intervening stretches of free DNA such as the activated PHO5promoter \[[@B5],[@B6]\]. Despite these consistencies, a possible caveat to using ChIP to evaluate nucleosome occupancy is that immunoprecipitation efficiency can depend on epitope accessibility. Rather than having low occupancies, genomic regions depleted in the H3 ChIP might be inaccessible as a result of association with large protein complexes in chromatin. To investigate this possibility, we examined a published chromatin fractionation dataset in which cross-linked chromatin fragments were subjected to phenol-chloroform extraction and DNA that partitioned into the aqueous phase was quantified by microarrays \[[@B15]\]. Given the polar nature of DNA and the hydrophobic nature of denatured protein, aqueous extraction should generally enrich for free DNA. We found that regions depleted in the H3 ChIP assay overlap extensively with regions enriched by aqueous extraction, but not with regions depleted by aqueous extraction (Figure [1d](#F1){ref-type="fig"}). Overall, there is a negative correlation of -0.54 between the H3 ChIP and aqueous-extraction datasets. Although the fractionation data may partially reflect differential cross-linking of lysines in the histone tails \[[@B15]\], this analysis suggests that regions depleted in the H3 ChIP experiment are relatively protein-free, as would be expected of non-nucleosomal DNA. Nucleosome occupancy correlates inversely with promoter strength ---------------------------------------------------------------- As previous studies show that PHO5activation is accompanied by marked nucleosome loss in the promoter region \[[@B5],[@B6]\], we sought to determine whether nucleosome depletion is a general attribute of active promoters. A total of 4,365 intergenic regions that reside immediately upstream of one or more validated yeast genes were assigned as promoters. Relative transcription rates were determined for each yeast gene from transcript levels measured by array and previously collected mRNA half-life data \[[@B16]\]. We found an inverse correlation of -0.39 between the enrichment of promoters in the H3 and H2B ChIP assays and the transcription rates of downstream genes (Figure [2a](#F2){ref-type="fig"}). Under the conditions examined, PHO5is not induced and its promoter has an average nucleosome occupancy according to these datasets. To evaluate further the relationship between nucleosome depletion and transcription, we collated a set of 308 nucleosome-depleted promoters on the basis of their relative depletion across the replicate H3 and H2B experiments. Of these nucleosome-depleted promoters, 42% regulate highly active genes (Figure [2b](#F2){ref-type="fig"}). These data suggest that there is a systematic relationship between promoter strength and nucleosome depletion. However, as this correspondence is not complete there are likely to be other determinants of nucleosome occupancy. Transcription factor binding motifs are over-represented in nucleosome-depleted promoters ----------------------------------------------------------------------------------------- To identify additional determinants of occupancy, we sought sequence elements associated with nucleosome depletion. Specifically, we carried out an unbiased search for elements up to 10 bp in length that occur with higher frequency in nucleosome-depleted promoters. Two distinct categories of sequences emerged (Figure [3a](#F3){ref-type="fig"}). The first includes poly(dA.dT) elements. Stretches of 10 or more dA.dT nucleotides appear in 38% of depleted promoters, compared with 26% of promoters overall (hypergeometric p\< 10^-5^). dA.dT stretches destabilize nucleosome formation in vitroand in vivo\[[@B17],[@B18]\]. The enrichment of poly(dA.dT) elements in nucleosome-depleted promoters probably reflects, at least in part, this destabilizing influence. As a high proportion of the poly(dA.dT) elements identified in nucleosome-depleted promoters are more than 10 bp long (30% are at least 14 bp), these data do not address the minimum length required for destabilization. However, in vitrostudies show that a 16-bp insertion leads to a 1.7-fold increase in accessibility of nucleosomal target sites \[[@B18]\]. The second sequence element enriched in nucleosome-depleted promoters corresponds to the consensus motif for the Rap1 transcription factor. This motif commonly occurs in the promoters of ribosomal proteins genes and is required for Rap1 binding in vitroand in vivo\[[@B19],[@B20]\]. Some variant of this motif appears in 22% of nucleosome-depleted promoters, compared with just 8% of promoters overall (hypergeometric p\< 10^-5^). Furthermore, multiple Rap1 sites are found in 19% of nucleosome-depleted promoters with Rap1 sites, compared to 8% of promoters with Rap1 sites overall (hypergeometric p\< 10^-3^). These data suggest that Rap1 recruitment may lead to nucleosome loss. Because only the Rap1 consensus site was identified in an unbiased search, we sought to identify additional sequence motifs by incorporating species conservation data. Specifically, we evaluated a set of 71 conserved motifs identified by Kellis and colleagues, a majority of which function in transcription factor recruitment \[[@B21]\]. Nearly half of these 71 motifs are over-represented in nucleosome-depleted promoters relative to promoters overall, as defined by a hypergeometric p\< 0.001. However, many of the implicated motifs appear in the same promoters. For example, nine of the over-represented motifs are associated with filamentation gene promoters \[[@B21]\]. We therefore considered the possibility that the total number of conserved motifs might be a more relevant predictor of nucleosome depletion. Indeed, we found that 31% of nucleosome-depleted promoters contain at least eight motifs, compared with 11% of promoters overall (hypergeometric p\<10^-5^; Figure [3b](#F3){ref-type="fig"}). Furthermore, nucleosome-depleted promoters contain an average of 6.1 motifs, whereas the average promoter contains 3.1 (permutation p\< 0.001; Figure [3c](#F3){ref-type="fig"}). Next, we sought motifs associated with nucleosome depletion in the absence of multiple motifs, by confining our analysis to promoters containing a maximum of four motifs. This analysis identified just two over-represented motifs, which correspond to the Rap1 and Swi4 binding sites. Hence, although a large number of conserved motifs are enriched in nucleosome-depleted promoters, most appear to be relevant mainly when occurring in combination. Functionally cooperative transcription factors associate with nucleosome-depleted promoters ------------------------------------------------------------------------------------------- As a majority of the conserved motifs recruit transcription factors \[[@B21]\], we examined the relationship between transcription factor binding and nucleosome occupancy more directly. Lee and colleagues combined ChIP and microarrays to identify target promoters for essentially all yeast transcription factors under the same conditions used here to evaluate nucleosome occupancy \[[@B1]\]. For each factor, we determined the significance of overlap between its target promoters and the set of nucleosome-depleted promoters. Of the 113 transcription factors in their database, 31 tend to associate with nucleosome-depleted promoters as defined by a hypergeometric p\< 0.001. Rap1 has the most significant association (Figure [4a](#F4){ref-type="fig"}), consistent with the enrichment of its binding motif (see above). Other top-ranked factors include Fhl1, which associates with many Rap1-bound promoters, and Swi4, whose binding motif is also enriched (Table [1](#T1){ref-type="table"}). We sought an underlying binding mechanism or function common to the transcription factors we had identified. However, these factors utilize a variety of binding domains, regulate different pathways, and only a minority have significant associations with promoters of highly active genes. Nonetheless, a commonality does emerge when transcription factor cooperativity is considered. A recent informatics study by Banerjee and Zhang identified 31 functionally cooperative transcription factor pairs (representing a total of 33 factors) on the basis of comprehensive binding and expression data \[[@B22]\]. Only a fraction of these are known to interact physically, suggesting that other mechanisms also confer cooperative function. There is a remarkable correspondence between these functionally cooperative factors and those that preferentially associate with nucleosome-depleted promoters (see Table [1](#T1){ref-type="table"}). Of the 31 factors we found to associate with nucleosome-depleted promoters, 17 were found to be functionally cooperative by Banerjee and Zhang (p\< 10^-5^). Furthermore, an evaluation of nucleosome occupancy at promoters bound by both members of a cooperative pair revealed a significant association with nucleosome-depletion for 18 of the 31 pairs (hypergeometric p\< 0.01). Together, these findings suggest that binding motifs and transcription factors act in combination to deplete nucleosomes and suggest a role for nucleosomes in transcription factor cooperativity \[[@B23]-[@B25]\]. Conditional nucleosome depletion at Rap1 consensus motifs --------------------------------------------------------- Although a number of transcription factors appear to act in defining promoter nucleosome occupancy, only the Rap1 consensus motif was identified in an unbiased search of nucleosome-depleted promoters. Furthermore, there is a highly significant association between nucleosome-depleted promoters and promoters bound by this factor in vivo\[[@B1]\] (Figure [4a](#F4){ref-type="fig"}). To investigate the relationship between Rap1 recruitment and nucleosome depletion further, we used ChIP and real-time PCR to evaluate nucleosome occupancy at several Rap1 binding sites in ribosomal protein promoters. We found that these regions are depleted 3- to 10-fold in H3 and FLAG-H2B ChIP assays, relative to a control promoter (TUB2) with average occupancy by global analysis (Figure [4b](#F4){ref-type="fig"}). We also used an orthogonal approach in which micrococcal nuclease digestion \[[@B26]\] was used to probe for nucleosomes at the TUB2, RPS11Band RPS15promoters (Figure [4c](#F4){ref-type="fig"}). A pattern of nuclease protection indicative of a regular nucleosome array is evident at the TUB2promoter, consistent with the average nucleosome occupancy attributed to this promoter by global ChIP analysis. In contrast, nuclease protection is not evident at the RAP1 sites in the RPS15promoter, consistent with the marked nucleosome-depletion attributed to this region by global ChIP and real-time PCR analysis. The region surrounding the RAP1 sites in RPS11Bexhibits weak nuclease protection, consistent with the modest nucleosome-depletion attributed to this region by global ChIP and real-time PCR. Although these focused analyses specifically addressed Rap1 sites in ribosomal protein genes, our global analyses indicate that approximately 30% of nucleosome-depleted promoters containing Rap1 motifs do not regulate ribosomal protein genes. Together these data confirm that nucleosomes are markedly depleted in the vicinity of Rap1 consensus sites in vivo, and thus extend previous studies showing that Rap1 induces local alterations in chromatin structure that, for example, result in increased nuclease sensitivity \[[@B27]-[@B29]\]. To gain further insight into the relationship between Rap1 and nucleosome depletion, we examined a mutant RPS11Bpromoter lacking its Rap1 consensus sites. We found that removal of these sites, which completely abrogates Rap1 binding \[[@B30]\], causes nucleosomes to return to the region, as reflected by a greater than twofold change in H3 ChIP enrichment (Figure [4d](#F4){ref-type="fig"}). We also examined the effect of rapamycin treatment on nucleosome occupancy in the vicinity of these consensus sites. Although ribosomal protein gene expression is dramatically reduced by rapamycin \[[@B31],[@B32]\], Rap1 remains bound to its target promoters (\[[@B30],[@B33]\], and B.B., E.P. and S.S., unpublished results). We found that rapamycin treatment causes nucleosomes to return to the vicinity of Rap1 sites, as reflected by twofold and greater increases in H3 ChIP enrichment (Figure [4e](#F4){ref-type="fig"}). Together these data show that Rap1 consensus sites are required for conditional nucleosome depletion at ribosomal protein gene promoters. Discussion ========== To gain further insight into the role of nucleosomes in gene regulation, we systematically evaluated promoter nucleosome occupancy in yeast by immunoprecipitating nucleosomal DNA and quantifying enrichment with microarrays. Promoters that are inefficiently immunoprecipitated by general anti-histone antibodies, and are therefore presumed to be relatively nucleosome-depleted, tend to regulate active genes (Figure [2](#F2){ref-type="fig"}). This is consistent with the previous observation that the activated PHO5promoter is largely devoid of nucleosomes \[[@B5],[@B6]\]. However, as not all nucleosome-depleted promoters regulate active genes, there are most likely to be additional determinants of depletion. An unbiased search for sequence elements enriched in nucleosome-depleted promoters revealed poly(dA.dT) elements, previously shown to destabilize nucleosome formation \[[@B17],[@B18]\], and the Rap1 consensus motif. By incorporating sequence conservation data \[[@B21]\], more than 30 other enriched motifs could be identified. However, most of these appear to be relevant mainly when occurring in combination. When we limited this analysis to promoters containing four or fewer motifs, all but two of these additional motifs drop out (only the Rap1 and Swi4 consensus sites remain). As the majority of conserved motifs incorporated in this analysis recruit transcription factors \[[@B21]\], these data suggest that multiple transcription factors act in combination to deplete nucleosomes. This possibility is further supported by our finding that functionally cooperative transcription factors tend to bind nucleosome-depleted promoters. These associations may reflect a mechanistic model in which transcription factors compete collaboratively to displace nucleosomes in order to gain access to target sites in the DNA \[[@B23]\]. This model was formulated to explain why certain pairs of transcription factors bind cooperatively to proximal target sites in vivoand on a chromatin template, but not to naked DNA \[[@B23]-[@B25]\]. This view invokes a broad role for nucleosomes as ubiquitous negative regulators of transcription factor binding and function. We speculate that by promoting synergy among multiple transcription factors and impeding the activities of individual ones, nucleosomes facilitate threshold behavior and filter noise (for example, genetic variation in motif sequence) in the transcriptional regulatory network. Although many factors appear to act in defining promoter nucleosome occupancy, our data indicate that Rap1 has a uniquely important role. Rap1 and its consensus motif are both markedly enriched in nucleosome-depleted promoters. Follow-up studies using real-time PCR and micrococcal nuclease digestion also demonstrate marked nucleosome depletion in the vicinity of Rap1 sites in the promoters of ribosomal protein genes. Moreover, nucleosomes appeared to return when the Rap1 consensus sites in one of these promoters were removed. These findings are consistent with previously described roles for Rap1 in opening chromatin and altering nucleosome positioning \[[@B27],[@B28]\]. However, Rap1 recruitment is not equally associated with nucleosome depletion under all conditions. We find that nucleosomes partially return to the vicinity of Rap1 sites during a rapamycin-induced starvation response \[[@B34]\], even though Rap1 remains bound (\[[@B30],[@B33]\], and B.B, E.P. and S.S., unpublished results). Hence, the nucleosome loss associated with Rap1 recruitment is most likely to require additional proteins, such as Esa1, a histone acetyltransferase recruited by Rap1 under exponential growth conditions but released in stress \[[@B30]\]. These findings may also offer insight into the barrier activity previously documented for Rap1 \[[@B35]\]. Heterochromatin propagation involves the sequential modification of histones in adjacent nucleosomes through positive-feedback mechanisms \[[@B7],[@B11]\]. Certain factors such as Rap1 are able to block this propagation by largely unknown mechanisms \[[@B36]\]. One model speculates that these barriers create nucleosome-free \'holes\' lacking the histone substrate required for heterochromatin propagation \[[@B29],[@B35]\]. By identifying such a \'hole\' in the vicinity of Rap1-binding sites in vivoour data support this model. Remarkably, the nucleosomal hole and the barrier function ascribed to Rap1 may be conditional, as nucleosomes return following treatment with the small molecule rapamycin, which activates a starvation response. Heterochromatic silencing has been shown previously to moderate under these conditions \[[@B37]\]. Hence, we speculate that dynamic influences on nucleosome occupancy may enable Rap1 to define chromatin domains and vary them in response to environmental cues. More broadly, the widespread nucleosome loss observed in the promoters of active genes provides a general caveat for ChIP studies examining posttranslational histone modifications, as a decrease in signal for a histone modification at a promoter undergoing activation may actually reflect nucleosome loss. Similarly, regions that appear relatively hypo-modified by ChIP may actually be nucleosome-depleted. However, this is not the case for low levels of acetylation \[[@B38]\] and H3 lysine 4 methylation \[[@B39]\] observed at yeast telomeres, as these regions have high occupancy. The data also provide insight into the maintenance of epigenetic information by histone modifications. Whereas epigenetic memory of a repressed state can be maintained on histones in promoters, memory of an activated state must be maintained on histones outside the promoters, for example in transcribed regions, which may not undergo significant nucleosome loss during activation \[[@B5],[@B6]\]. Methylation of histone H3 at lysines 4 and 36, targeted to transcribed regions in yeast via interactions between RNA polymerase and the methylases \[[@B39]-[@B47]\], may represent such \'activating\' marks. Materials and methods ===================== Chromatin immunoprecipitation (ChIP) ------------------------------------ DNA associated with histone H3 in vivowas immunoprecipitated with antibodies against the invariant H3 carboxy terminus using a ChIP protocol described previously \[[@B39],[@B48],[@B49]\]. Briefly, 45 ml log-phase w303a yeast (OD~600~\~ 1.0) growing in yeast extract/peptone/dextrose (YPD) were cross-linked in 1% formaldehyde for 15 min, washed twice in PBS, resuspended in 400 μl lysis buffer (50 mM Hepes-KOH pH 7.5, 140 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate) and lysed with glass beads. The resulting extract was sonicated to fragment chromatin (4 × 20 sec burst/30 sec rest with a Branson Sonifier 250 at 70% duty, power 3) and centrifuged for 15 min. Solubilized chromatin was then immunoprecipitated with polyclonal antibodies against the carboxy terminus of histone H3 (Abcam or Cell Signaling). A unenriched whole-cell extract sample (WCE) was also retained as a control. After enrichment, cross-links were reversed by incubating samples in 10 mM Tris-HCl pH 8.0, 1 mM EDTA, 1.0% SDS, 150 mM NaCl at 65°C overnight. DNA was purified from ChIP and WCE samples by proteinase K treatment, phenol/chloroform extraction, ethanol precipitation, and incubation with RNAse. DNA associated with histone H2B in vivowas isolated in a similar manner from yeast containing epitope-tagged H2B \[[@B50]\] using anti-FLAG M2 monoclonal antibodies (Sigma). DNA amplification and hybridization ----------------------------------- To obtain sufficient quantities for hybridization, immunoprecipitated DNA (from approximately 10^8^cells) and whole-cell extract DNA (unenriched control) were amplified in a linear fashion as described \[[@B51]\]. Briefly, terminal transferase was used to add poly(T) tails to DNA fragment and a T7-poly(A) adaptor primer was used to incorporate T7 promoters. The reaction products were used as template for an in vitrotranscription reaction carried out with the T7 Megascript Kit (Ambion) and RNA samples were purified using an RNeasy Mini Kit (Qiagen). Amplified RNA was reverse-transcribed, incorporating amino-allyl dUTP, and the resulting DNA was fluorescently labeled by incubation with monofunctional reactive Cy5 (enriched sample) or Cy3 (unenriched control) dye as described \[[@B52]\]. Microarrays containing 6,438 PCR-amplified intergenic regions were prepared as described previously \[[@B39],[@B53],[@B54]\]. Mixed Cy5-/Cy3-labeled probe was hybridized to intergenic microarrays for 12-14 h at 60°C, washed and then scanned using a GenePix 4000A scanner with GenePix Pro software (Axon Instruments) as described \[[@B55]\]. In addition, transcript levels were determined by hybridizing Cy5-labeled mRNA extracted from log phase w303a yeast against Cy3-labeled genomic DNA on microarrays containing 6,218 open reading frames (ORFs), as described previously \[[@B16]\]. Microarray data processing -------------------------- Cy5 and Cy3 fluorescence were integrated for each feature using GenePix Pro Software (Axon). Data were processed and composite Cy5:Cy3 ratios determined according to protocols at the Stanford Microarray Database \[[@B56]\]. Correlations between replicate datasets were \~0.8 for all experiments. Composite datasets were log~2~transformed and zero centered before further analysis. The histone H3 ChIP dataset was determined from four independent immunoprecipitations and hybridizations (two each using antibodies from Cell Signaling or Abcam). The FLAG-H2B ChIP dataset was determined from three independent immunoprecipitations and hybridizations. The mRNA dataset was determined from three independent extractions and hybridizations of mRNA against genomic DNA. Relative transcription rates were determined by dividing transcript levels by half-life data collected by Wang and colleagues \[[@B16]\]. A set of activated promoters was defined as those in the top 10% by mRNA expression level of associated gene, with divergent promoters assigned to the more highly expressed gene. Complete datasets are available online \[[@B57]\]. Analysis of nucleosome-depleted promoters ----------------------------------------- Z-scores were assigned to each intergenic that reflect depletion across the four H3 and three H2B ChIP experiments, using the formula Z = (x - μ)/σ where x is the average of the replicate measurements, μ is the average of all intergenics and σ is the standard error of the replicate measurements. We defined as nucleosome-depleted the 410 features with the highest Z-scores. This set, which includes 308 promoters, contains nucleosome-depleted outliers and is not inclusive of all promoters that immunoprecipitate with average or lower efficiency. The average aqueous enrichment ratio \[[@B15]\] for these 308 depleted promoters is 1.7-fold, significantly higher than expected by chance (permutation p\< 0.001), consistent with the premise that these promoters are relatively free of nucleosomes. Sequence elements common to nucleosome-depleted promoters were identified by searching between 10 and 500 bp upstream of gene start sites for over-represented sequences up to 10 bp in length using the GeneSpring program suite (Silicon Genetics). Enrichment was confirmed by evaluating the significance of overlap between the set of nucleosome-depleted promoters and the set of promoters containing Rap1 consensus motifs (ACACCCATACAT with up to two mismatches) or poly dA.dT stretches at least 10 bp in length (identified using PatMatch, SaccharomycesGenome Database \[[@B58]\]). Statistical significances of overlaps between sets are expressed as P-values calculated by a hypergeometric probability model. The P-values reflect the extent to which observed overlaps exceed that expected under the null hypothesis that there is no relationship between the sets \[[@B59]\]. Where specified, permutation analyses were carried out by generating 1,000 random but representative promoter sets with an Excel macro and used to confirm statistical significance. Lists of promoters containing the 71 conserved motifs \[[@B21]\] were collated from gene sets available online \[[@B60]\]. Lists of promoters bound by transcription factors at a significance of p\< 0.001 \[[@B1]\] were collated from data available at \[[@B61]\]. Real-time PCR ------------- Regions approximately 200 bp in size that span one or more Rap1 consensus sites in ribosomal protein gene promoters were amplified from ChIP and unenriched control samples using SYBR green PCR mix (Qiagen) in an MJ Research real-time PCR machine according to the manufacturers\' instructions. Fold-ratios that reflect relative enrichment or depletion of a given region in the H3 or FLAG-H2B ChIP assays were determined using the 2^-ΔΔC^~T~method described in the Applied Biosystems User Bulletin. For each region examined, the TUB2promoter was used as the normalizer (this promoter is used as a control because its occupancy approximates that of the average promoter by global analysis), and the unenriched control sample was used as the calibrator. Each reported ratio represents the average of three independent ChIP experiments analyzed in duplicate by real-time PCR. The following primer pairs were used: RPS22A promoter:5\'-GCCTAAAACGCCCATAAGTT-3\' and 5\'-ACTGCAAACCCATATTCAAGA-3\' RPS15 promoter:5\'-TACACCGCGCGTATAAATCA-3\' and 5\'-CCCAGCAAGGAGTTTCTCAG-3\' RPS11B promoter:5\'-GAAGAAATATTTCCTTGCTGCACC-3\' and 5\'-AAGGGAAACGTAAAGCTATTGGAC-3\' RPL23A promoter:5\'-ATTAACATCTGTACACCCCCAACT-3\' and 5\'-TACAGTTCGTTTCCTGCC ATATTA-3\' TUB2 promoter:5\'-GGCCTAACAGTAAAGATATCCTCC-3\' and 5\'-GTTGTAGTAGCTGCTATGT CACTC-3\' Centromeric vectors containing either a mutant RPS11Bpromoter lacking the two Rap1 consensus motifs \[[@B30]\] or an essentially wild-type allele were transformed into wild-type yeast and used in an H3 ChIP assay to evaluate the consequence of removing Rap1 binding sites on nucleosome occupancy. Enrichment was evaluated by real-time PCR using the following primer pair that selectively amplifies the plasmid alleles but not the endogenous RPS11Bpromoter: 5\'-CTGGAAGAAATATTTCCTT GCTCTAG-3\' and 5\'-AAGGGAAACGTAAAGCTATTGGAC-3\'. Micrococcal nuclease assay -------------------------- Log-phase cultures of W303a yeast grown in 450 ml YPD to OD~600~of 1.0 were spheroplasted with zymolase (10 mg in 40 ml volume of 1 M sorbitol, 50 mM Tris pH 7.4, 10 mM β-mercaptoethanol (β-ME), at 30°C for 38 min shaking at 300 rpm), divided into five aliquots, and digested with increasing concentrations (20 U to 320 U) of micrococcal nuclease (Worthington Biochem) in 600 μl 0.5 mM spermidine, 1 mM β-ME, 0.075% NP-40. DNA from digested samples was extracted with phenol twice and chloroform once and precipitated in ethanol. Samples were washed, resuspended in 10 mM Tris pH 7.5, subjected to RNAse treatment, cleaned up with the MinElute kit (Qiagen) and run out in a 1% agarose gel. Following depurination, denaturation and neutralization of the gel, DNA was transferred onto nylon membranes by capillary action and covalently linked to the membranes by UV irradiation. Southern blotting was carried out using a DIG Luminescent Detection Kit (Roche) and DIG-labeled probe generated by PCR using the TUB2, RPS11Band RPS15primers described above. Acknowledgements ================ We thank Mary Ann Osley and Kevin Struhl for generously providing yeast strains, and Jay Bradner, Jeff McMahon, Aly Shamji, Jianping Cui, Manolis Kellis, Vamsi Mootha, Mike Kamal and Oliver Rando for insightful discussions. This study was supported by a grant from NIGMS (GM38627, awarded to S.L.S.). B.E.B. is supported by a K08 Development Award from the National Cancer Institute. C.L.L. is supported by a Graduate Research Fellowship from the National Science Foundation. S.L.S. is an Investigator at the Howard Hughes Medical Institute. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Correlation between H3 and FLAG-H2B ChIP datasets. DNA associated with histones in vivowas enriched in ChIP assays using antibodies against histone H3 or FLAG-H2B, and quantified by microarrays. (a)Relative enrichment of promoters and other non-coding regions in the H3 and H2B ChIP assays is shown. (b)Histogram showing distributions of enrichment for promoter regions in the H3 and H2B ChIP assays. (c)Overlap between regions depleted in the H3 and FLAG-H2B assays is shown. Overall, there is an 0.83 correlation between these ChIP datasets. (d)Overlap between regions depleted in the H3 ChIP assay and regions enriched by aqueous extraction is shown \[62\]. ::: ![](gb-2004-5-9-r62-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Inverse association between nucleosome occupancy and promoter strength. (a)Relative enrichment of promoter regions in the H3 and FLAG-H2B ChIP assays plotted against transcription rate of downstream genes (moving average, window 50). (b)Overlap between promoters upstream of active genes and the set of nucleosome-depleted promoters defined on the basis of depletion across the replicate H3 and FLAG-H2B experiments. ::: ![](gb-2004-5-9-r62-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Sequence motifs over-represented in nucleosome-depleted promoters. (a)An unbiased search for sequences up to 10 bp in length over-represented in nucleosome-depleted promoters (relative to promoters overall) identified the poly(dA.dT) sequence element and variants of the Rap1 consensus motif ACACCCATACAT \[21\]. (b)Overlap between nucleosome-depleted promoters and promoters that contain multiple conserved motifs is shown \[21\]. (c)Histogram showing average numbers of motifs in 1,000 randomly generated promoter sets. Nucleosome-depleted promoters contain an average of 6.1 conserved motifs, significantly higher than in these randomly generated sets. ::: ![](gb-2004-5-9-r62-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Nucleosome depletion in the vicinity of Rap1-binding sites. (a)Overlap between the 308 most nucleosome-depleted promoters and promoters found to recruit Rap1 in a global ChIP study \[1\]. (b)Nucleosome depletion in the vicinity of Rap1-binding sites in ribosomal gene promoters evaluated by ChIP. Fold-enrichment was determined by real-time PCR using primers that span Rap1-binding motifs in the RPS22A, RPS15, RPS11Band RPL23Apromoters. (c)Southern blots showing DNA from yeast spheroplasts digested with increasing concentrations of micrococcal nuclease probed with labeled PCR products spanning the TUB2promoter and the Rap1 sites in the RPS11Band RPS15promoters. (d)Nucleosome occupancy for a mutant RPS11Bpromoter lacking Rap1 consensus sites was determined by H3 ChIP and real-time PCR. The mutant promoter is enriched 2.1-fold relative to wild type. (e)Nucleosome occupancy at Rap1-binding sites in ribosomal protein gene promoters after treatment with rapamycin evaluated by H3 ChIP and real-time PCR. ::: ![](gb-2004-5-9-r62-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Transcription factors that tend to associate with nucleosome-depleted promoters ::: Transcription factor Pathway Number of targets Nucleosome-depleted Functionally cooperative ---------------------- ------------------------ ------------------- --------------------- -------------------------- Rap1 Biosynthesis 291 35% Fhl1 Biosynthesis 137 48% Swi4 Cell cycle 165 36% Hsf1 Environmental response 114 35% Gat3 Metabolism 119 31% Cin5 Environmental response 200 23% Phd1 Metabolism 138 25% Dal81 Metabolism 70 34% Ndd1 Cell cycle 122 26% Yap6 Environmental response 123 26% Fkh2 Cell cycle 145 24% Pdr1 Environmental response 103 27% Ino4 Metabolism 118 25% Smp1 Environmental response 99 27% Yap5 Environmental response 113 26% Ash1 Development 41 41% Transcription factors are ranked according to the significance of their association with nucleosome-depleted promoters, as determined by a hypergeometric model. Shown are the 16 top-ranked factors along with relevant physiologic pathway, number of promoters bound \[1\], and percent of target promoters that are nucleosome-depleted. Factors found previously to be functionally cooperative are indicated \[22\]. :::	PubMed Central	2024-06-05T03:55:48.490820	2004-8-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522869/", "journal": "Genome Biol. 2004 Aug 20; 5(9):R62", "authors": [ { "first": "Bradley E", "last": "Bernstein" }, { "first": "Chih Long", "last": "Liu" }, { "first": "Emily L", "last": "Humphrey" }, { "first": "Ethan O", "last": "Perlstein" }, { "first": "Stuart L", "last": "Schreiber" } ] }
PMC522870	Background ========== Physical interactions between proteins underpin most biological processes. For this reason, large-scale protein-interaction mapping projects have been initiated in several model organisms \[[@B1]-[@B6]\]. Unfortunately, projects of a similar scale have not yet been described for mammalian systems, with the result that our global understanding of protein function remains less advanced in mammals than in lower eukaryotes. However, many physical interactions are conserved between species, so it should be possible to infer information about human protein interactions and protein function using data from model organism protein-interaction datasets \[[@B7],[@B8]\]. To transfer information on gene function between two genomes requires the identification of orthologous genes in the two genomes (that is, genes that are descended from a common ancestor and share biological functions). However, the identification of gene orthologs is often not a trivial problem; gene duplications can result in a single gene having multiple potential orthologs in a second species. In addition, it is necessary to distinguish true gene orthologs from \'out-paralogs\' (that is, genes that arose from a gene-duplication event before the divergence of two species, and so are unlikely to share functions) \[[@B9]\]. One method that addresses both these problems is the InParanoid algorithm, which first identifies potential orthologs by best pairwise similarity searches, and then clusters these orthologs into groups of likely co-orthologs, with each ortholog assigned a score representing the confidence that it is the main ortholog \[[@B9]\]. We have used the orthology relationships identified by the InParanoid algorithm to construct a putative human protein-interaction map based solely on high-throughput interaction datasets from model organisms. We show that this approach successfully identifies functionally related human proteins, and so can be used to assign putative functions to many novel human genes. The resulting network provides a framework for human biology and acts as a guide for a future experimental human protein-interaction mapping project. Results ======= Generation of a human protein-interaction map --------------------------------------------- Protein interactions are often evolutionarily conserved between orthologous proteins from different species \[[@B7]\]. Hence we reasoned that a human protein-interaction map could be constructed using data from model organism protein-interaction mapping projects. We obtained the data from seven experimental and four computationally predicted protein-interaction maps from Saccharomyces cerevisiae\[[@B1]-[@B4],[@B10],[@B11]\], Drosophila melanogaster\[[@B5]\] and Caenorhabditis elegans\[[@B6]\]. For each interacting protein, we identified potential human orthologs using the InParanoid algorithm \[[@B9]\]. A human protein interaction is predicted if both interaction partners from a model organism have one or more human orthologs. Using this strategy, we were able to generate a human interaction network comprising 71,496 interactions between 6,231 human proteins. The sources of these predicted interactions are summarized in Table [1](#T1){ref-type="table"} and Figure [1a](#F1){ref-type="fig"}, and all the interactions are available in Additional data file 1 available online with this article and can also be searched or downloaded from our website \[[@B12]\]. Assessment of the accuracy of the interaction datasets ------------------------------------------------------ In the absence of a comprehensive set of verified human protein interactions, we required another method to assess the accuracy of the interaction network. Proteins that interact physiologically are expected to have related functions. Therefore high-quality interaction datasets should predict a greater proportion of interactions between functionally related proteins than low quality datasets. The functions of human proteins can be systematically described using the Gene Ontology (GO) annotations \[[@B13]\] available from Ensembl \[[@B14]-[@B17]\]. GO annotations provide a hierarchical description of gene functions with general functions described by GO annotations at the top levels of the hierarchy and very precise functions described by terms deeper in the hierarchy. Because physiologically interacting proteins are expected to have related, but non-identical functions, they are expected to share some, but not all GO annotations. Therefore, one method to evaluate an interaction dataset is to count the proportion of interactions that connect proteins that share common GO terms \[[@B5]\]. For the complete predicted human interaction network, 25% of interaction partners share at least one GO term, which is many more than observed with a randomly generated network of the same size (15% of interactions). To confirm that this result did not just apply to quite general GO annotations, we calculated the proportion of interaction partners that share GO annotations at depths 3 to 8 and greater than 8 in the GO hierarchy. We found that the predicted interaction network preferentially connects proteins that share GO annotations at any level of the GO hierarchy (see Figure [2](#F2){ref-type="fig"}). This suggests that the interaction network indeed preferentially connects functionally related human proteins. We then used the same strategy to compare the accuracy of human interactions predicted by data from the three different model organisms. If the interactions from a particular model organism dataset predict fewer interactions between functionally related human proteins than the other datasets, then this dataset should be considered less reliable as a source of candidate human protein interactions. As shown in Table [1](#T1){ref-type="table"} and Figure [2a](#F2){ref-type="fig"}, interactions predicted by the complete yeast and worm datasets are slightly better at connecting functionally related human proteins than those predicted by the fly dataset, suggesting that these interactions can be considered with higher confidence. This result is especially interesting given that the yeast interaction map is an order of magnitude larger than the fly or worm maps, confirming that the fly and worm interaction maps currently have a relatively low coverage. Next we asked how the confidence in the assignment of gene orthologs affects the accuracy of an interaction. For each predicted interaction, an orthology confidence score was calculated by summing the InParanoid orthology confidence scores for the two human and two model organism proteins (see Materials and methods). Of the predicted interactions, 24,897 have the maximum possible confidence score of 4. Of these interactions, 28%, 24% and 13% connect proteins that share GO terms at depths of 3, 5 or 7 in the GO hierarchy (excluding proteins without GO annotation). In contrast, for interactions with an orthology confidence score less than 4, these figures are 24%, 20% and 10%. Hence we conclude that the predicted human interactions with high-confidence orthology assignments can be considered more reliable than those interactions with less confidence in their orthology assignments. This confirms that the confidence scores assigned using InParanoid are indeed likely to be useful predictors of functional conservation. A core dataset of high-confidence protein interactions ------------------------------------------------------ The worm and fly interaction mapping projects both defined a subset of high-confidence \'core\' interactions that have the greatest experimental support (Figure [1b](#F1){ref-type="fig"}). For the worm interaction map these were defined as interactions identified more than once, or that reconfirmed when retested in the two-hybrid assay \[[@B6]\]. In the fly interaction map each interaction has an associated confidence score, and interactions with a score greater than 0.5 are considered core interactions (the interaction score mainly depends upon the number of times each interaction was detected, the total number of interactions made by each protein and the local network clustering; see \[[@B5]\]). To generate a similar subset of yeast protein interactions, we defined core yeast protein interactions as those identified more than once by any single assay, consistent with previous analyses of the individual datasets \[[@B1]-[@B3],[@B11]\]. As shown in Figure [2a](#F2){ref-type="fig"} and Table [1](#T1){ref-type="table"}, for all three species these core interactions predict a greater proportion of human interactions that share GO terms than the total datasets. Indeed all three core interaction maps are of similar accuracy, so we combine their predicted interactions into a core network of 11,487 higher-confidence human protein interactions (summarized in Table [2](#T2){ref-type="table"} and available as Additional data file 2). Of these core interactions, 38%, 35% and 24% connect proteins that share GO terms at depths of 3, 5 or 7 in the GO hierarchy (excluding proteins with no GO annotations). Combining interaction datasets to generate high-confidence networks ------------------------------------------------------------------- It has been shown previously that protein interactions detected by more than one high-throughput interaction assay are more accurate \[[@B11]\]. We find that this is also true for human protein interactions predicted by yeast protein interactions detected by more than one method (see Figure [2b](#F2){ref-type="fig"} and Table [1](#T1){ref-type="table"}). It has also been suggested that protein interactions are more likely to represent physiologically important interactions if they have been detected between orthologous protein pairs from two or more species \[[@B7],[@B18]\]. To test this hypothesis we identified 288 human protein interactions predicted by interactions in two or more model organisms (Figure [1](#F1){ref-type="fig"}, Table [1](#T1){ref-type="table"}). Remarkably, 75%, 70% and 56% of these interactions share GO terms at depths of 3, 5 or 7 in the GO hierarchy, respectively (Figure [2b](#F2){ref-type="fig"}). Indeed, for interactions derived from core interaction datasets, these figures rise to 88%, 80% and 67% of interactions. Hence, protein interactions predicted by data from multiple species can be considered with very high confidence. Using the interaction network to predict human gene function ------------------------------------------------------------ Because physiologically interacting proteins often have similar functions (Figure [2](#F2){ref-type="fig"}), it should be possible to predict the functions of a novel human protein if it interacts with proteins of known function. To address how well our interaction map could be used for this purpose, we asked whether the known GO terms of a protein could be predicted using only the GO terms of its interaction partners. As shown in Table [3](#T3){ref-type="table"}, GO terms associated with at least one of a gene\'s core interaction partners predict GO terms associated with that gene with an accuracy of around 8%. However, GO terms associated with at least two, three, four or five of a gene\'s interaction partners have 22%, 30%, 37%, 42% and 45% probabilities, respectively, of also being associated with that gene (Table [3](#T3){ref-type="table"}). Although these values may vary for different GO terms, as shown in Additional data file 3, the accuracy and coverage of these GO term predictions are very similar for GO terms at different levels in the GO hierarchy, and so can be used as an approximate indication of the confidence in a prediction of gene function. Hence the network can be used to predict GO terms for a human gene of unknown function, with the approximate confidence in the GO prediction determined by the number of interaction partners that share the GO term. The ability to provide a reasonably accurate prediction of a gene\'s GO terms means that we can use the interaction network to provide probabilistic gene function predictions for novel human proteins and also to predict additional functions for proteins with some known functions. The core interaction map contains 864 proteins with no functional annotations. About 10% of these proteins interact with two or more proteins that share GO terms. The probabilistic predictions of the functions of these novel proteins are listed in Additional data file 4. Often these predicted functions are also supported by the known functions of the protein domains predicted to be encoded by these novel genes (see Additional data file 4). For example, ENSG00000028310 encodes a bromodomain and interacts with six proteins annotated as \'GO:0006355 regulation of transcription, DNA-dependent\', ENSG00000080608 encodes an RNA-binding domain and interacts with five proteins annotated as \'GO:0006364 rRNA processing\', and ENSG00000104863 encodes a PDZ domain and interacts with three proteins with the annotations \'GO:0005887 integral to plasma membrane, GO:0007242 intracellular signaling cascade\' (Additional data file 4). The complete and core interaction maps also predict interactions for 448 and 292 human disease genes (listed in Additional data file 5), of which 55 interact with two or more proteins in the core interaction network that share a GO annotation. The functional predictions for these 55 genes are listed in Additional data file 6. Discussion ========== A framework for human biology ----------------------------- We report here the use of data from model organism protein-interaction mapping projects to predict a network of human protein interactions. This network consists of over 70,000 interactions that connect over one-third of all the predicted human proteins, including 1,482 proteins of unknown function and 448 proteins encoded by human disease genes. The physiological relevance of this network is supported by its ability to preferentially connect human proteins that share biological functions (Figure [2](#F2){ref-type="fig"}). Indeed the network can be successfully used to predict the functions of a gene using the known functions of its interaction partners (Table [3](#T3){ref-type="table"}). As such, the network should provide a rich source of functional hypotheses for researchers interested in the functions of one or many human proteins. The accuracy and coverage of the interactions predicted in this network depend primarily on two parameters: the quality of the original model organism interaction datasets; and the ability to identify the human orthologs of a model organism protein. Our analysis suggests that the raw yeast and worm protein-interaction datasets are currently slightly more accurate than the raw fly interaction dataset, but that when filtered for high-confidence interactions the three interaction maps are of very similar accuracy (see Table [1](#T1){ref-type="table"} and Figure [2](#F2){ref-type="fig"}). The fly and worm interaction maps both have a much lower coverage than the yeast interaction network, most probably because they both only represent the results of a single interaction-mapping project. The continuation of these model organism protein-interaction mapping projects to generate higher coverage interaction maps will greatly enhance our ability to predict human protein interactions. For the identification of gene orthologs, we used the InParanoid algorithm. InParanoid offers several important benefits compared to simple \'reciprocal best hit\' sequence-similarity searches \[[@B9]\]. First, many genes from lower eukaryotes have multiple co-orthologs in humans, which can be identified using InParanoid, but not by simple one-to-one sequence-similarity searches. Second, InParanoid can successfully distinguish these true co-orthologs from paralogs that arose before a speciation event (which are unlikely to retain similar functions). Finally, each potential ortholog in a group of co-orthologs identified by InParanoid has an associated score that represents the likelihood that it is the main ortholog of a gene. We have summed these confidence scores to provide an orthology confidence score for each predicted human protein interaction in our network. These high-confidence ortholog interactions connect a greater proportion of functionally related human proteins, suggesting that the InParanoid confidence score is indeed a useful tool for predicting the likely physiological relevance of a predicted protein interaction. The ability to successfully predict human protein functions using the results of model organism protein-interaction mapping projects highlights both the relevance of model organism protein-interaction mapping projects to understanding human biology and also the benefits that would result from an experimental human protein-interaction mapping project. Although the interaction network can currently accurately predict only a subset of the known functions of a gene, this should improve as more protein-interaction data becomes available. For this reason, we strongly encourage the continuation of model organism protein-interaction mapping projects. Methods of verifying protein-interaction datasets ------------------------------------------------- We also assessed the relative merits of three different methods to improve the accuracy of protein-interaction maps. The first strategy is to define a subset of interactions detected more than once with a single assay \[[@B1]-[@B3],[@B6]\]. We found that this approach leads to an approximately 1.5- to 2.7-fold increase in the proportion of predicted human interactions that share GO terms (Figure [2b](#F2){ref-type="fig"}). The second strategy is to define a subset of interactions that have been identified by more than one interaction assay. This results in around a 2.3- to 8-fold improvement in the prediction of associations between proteins that share GO terms (Figure [2b](#F2){ref-type="fig"}). The final strategy is to define a subset of interactions that are predicted by interactions from more than one model organism, which results in around a 3- to 12-fold improvement in the proportion of interactions between proteins sharing GO terms (Figure [2b](#F2){ref-type="fig"}). With all these filtering methods, the greatest improvements are seen when considering the proportion of interactions that share GO terms deep within the GO hierarchy; that is, the filtering steps dramatically improve the proportion of interactions between proteins with very closely related functions. We conclude that using interaction data derived from a second interaction assay or from a second species both represent excellent methods to improve the accuracy of protein-interaction maps. Because of the small number of protein-interaction assays that have been adapted to a high-throughput format, we suggest that constructing a second interaction map in a related organism using the same assay may be an efficient way to produce a high-confidence interaction map. This strategy is somewhat similar to using phylogenetic footprinting to identify functional noncoding DNA, so we suggest it should be named \'interaction footprinting\'. Using the relatively low-coverage model organism interaction datasets currently available, only a small proportion of interactions can be verified by interaction footprinting. The continuation of these model organism interaction mapping projects will not only provide a much richer framework of predicted human protein interactions, but will also allow many more interactions to be verified using the interaction footprinting strategy. However, such an approach will be limited to providing information on those proteins and interactions that are conserved between vertebrates and invertebrates. Strategies for completing the human interaction map --------------------------------------------------- The interactions described here provide a first-draft human protein-interaction map that can be used to predict interactions and functions for genes of interest to a particular researcher. However, the map also provides a framework from which a complete human protein-interaction map could be generated. Firstly, the map could be used to identify subsets of high-confidence, evolutionarily conserved interactions from the results of large- or medium-scale human interaction-mapping projects. For example the map verifies 51 of 296 yeast two-hybrid interactions detected for human proteins involved in mRNA decay \[[@B19]\]. Alternatively, the interactions predicted here could be directly experimentally validated using an assay that allows rapid testing of binary interactions (such as the yeast or mammalian two-hybrid assays \[[@B20]\] or protein fragment complementation assays \[[@B21]\]). This would represent a cost-effective strategy to produce a high-confidence human protein-interaction map because it massively reduces the number of candidate interactions that need to be tested. Finally, the map identifies 17,300 (23,531 - 6,231) human genes for which no protein interactions are predicted from model organism interaction datasets. Many of these proteins are likely to be vertebrate- or mammalian-specific, and are the most logical choices for bait proteins for the discovery phase of an experimental human protein-interaction mapping project. Materials and methods ===================== Model organism protein-interaction datasets ------------------------------------------- The interaction datasets used to generate the draft human protein-interaction network were two-hybrid-based interaction maps for D. melanogaster\[[@B5]\] and C. elegans\[[@B6]\] and a list of S. cerevisiaeprotein-interactions compiled by Von Mering et al.\[[@B11]\] from two two-hybrid \[[@B1],[@B2]\], two complex purification \[[@B3],[@B4]\], one genetic \[[@B10]\], and four in silico-predicted interaction datasets (which used correlated mRNA expressions, conserved gene neighbourhood, gene co-occurrence or gene fusion events to predict protein interactions \[[@B11]\]). Table [4](#T4){ref-type="table"} shows the number of unique interactions in each dataset, the methods used to generate each dataset, and the URLs from which the datasets were obtained. Identification of gene orthologs and construction of the interaction network ---------------------------------------------------------------------------- The human orthologs of yeast, worm and fly genes were identified using the InParanoid algorithm, which is designed to distinguish true orthologs from out-paralogs that arose from gene duplications before the divergence of two species \[[@B9]\]. The InParanoid algorithm first identifies potential orthologs by best pairwise similarity searches, and then clusters these orthologs into groups of probable co-orthologs, with each ortholog assigned a score representing the confidence that it is the main ortholog. For each interaction data source, we obtained SWISS-PROT/TrEMBL accessions for each interacting protein using the Ensmart data-mining tool \[[@B16],[@B17]\] (for worm and fly genes) or both SWISS-PROT \[[@B22]\] and a TrEMBL conversion file kindly provided by Paul Kersey, EBI, Hinxton, UK (for yeast genes). Potential human orthologs of these genes were then identified using the pre-computed InParanoid results (version 2.3, available from \[[@B23]\]), and the results converted to nonredundant Ensembl (v19.34a.1, genome assembly NCBI34) gene IDs using Ensmart (v19.1) 1 \[[@B16],[@B17]\]. In total, InParanoid identifies 9,500 human genes with at least one ortholog in at least one of worm, fly or yeast. For each potential ortholog in a group of co-orthologs, the InParanoid algorithm calculates a score that represents the confidence that it is the main ortholog. In this scoring system, the main ortholog always receives a score of 1, with the other co-orthologs receiving scores ranging between 0 and 1, calculated according to their similarity to the main ortholog \[[@B9]\]. As an indication of the confidence we have in the orthology relationships between a pair of interacting proteins from a model organism and a predicted pair of interacting human proteins, we calculate a confidence score by summing the InParanoid confidence scores for each of the four proteins. Hence, each interaction has an associated score ranging from 0 to 4 that represents the confidence that both human proteins represent the main orthologs of the model organism proteins, and vice versa. Core interactions were defined as those predicted by worm interactions identified more than once or that reconfirmed when retested in the two-hybrid assay \[[@B6]\], by fly interactions with an interaction score greater than 0.5 \[[@B5]\], or by yeast interactions detected two or more times by a single assay \[[@B1]-[@B3],[@B11]\]. Assessment of the interaction data ---------------------------------- Human GOs (at levels 3 or deeper in the GO hierarchy) were obtained from Ensembl (v19.34a.1) \[[@B14],[@B15]\] using Ensmart (v19.1) \[[@B16],[@B17]\]. The GO terms \'unknown molecular function/biological process/cellular compartment\' were discarded in all subsequent analyses. To validate the accuracy of the interaction data, we calculated the percentage of interactions that shared at least one GO term. To confirm that the results did not just apply to very general GO annotations, we calculated the proportion of interacting proteins that shared a GO annotation at levels 3 to 8 and greater than 8 in the GO hierarchy. For all of these analyses we ignored proteins with no associated GO annotations. Moreover, self-interactions were excluded because they will always share GO terms and so bias the results. Prediction of gene functions ---------------------------- To predict the GO terms of a protein, we identified all the GO terms associated with xor more of its interaction partners (where xvaried from 1 to 6). To validate the accuracy and coverage of this approach we predicted GO terms for genes that already have associated GO terms. The accuracy was calculated as the total number of correct GO term predictions divided by the total number of GO term predictions. The coverage was calculated as the total number of correct GO term predictions divided by the total number of known GO terms. This analysis was repeated, but only considering individually GO terms at depths of 3 to 8 and greater than 8 in the GO hierarchy (see Additional data file 3). To avoid biasing the results we again ignored self-interactions. For the same reason, we also only counted once GO terms associated with more than one interaction partner predicted by the same source interaction from a model organism. The InterPro protein domains \[[@B24]\] encoded by each human gene were obtained from Ensembl using Ensmart. Genes of unknown function were defined as those having no associated GO terms, and disease genes were as defined by Ensembl using the Online Mendelian Inheritance in Man (OMIM) database as a reference \[[@B25]\]. Additional data files ===================== The following additional data files are available with the online version of this article: Additional data file [1](#s1){ref-type="supplementary-material"} contains a complete list of predicted human protein interactions; this dataset contains every human protein interaction that is predicted by a protein interaction from any of seven experimental and four computationally-predicted protein interaction maps from Saccharomyces cerevisiae\[[@B1]-[@B4],[@B10],[@B11]\], Drosophila melanogaster\[[@B5]\] and Caenorhabditis elegans\[[@B6]\]. Additional data file [2](#s2){ref-type="supplementary-material"} contains a list of all core human protein interactions. This represents a subset of high-confidence human protein interactions that is predicted by model organism protein interactions with greater experimental support. In the worm interaction map, these are defined as interactions that reconfirmed when retested in the Y2H assay \[[@B6]\]. In the fly interaction map, each interaction has an associated confidence score, and interactions with a score greater than 0.5 are considered core interactions (the interaction score mainly depends upon the number of times each interaction was detected, the total number of interactions made by each protein and the local network clustering \[[@B5]\]). To generate a similar subset of yeast protein interactions, we defined core yeast protein interactions as those identified more than once by any single assay. Each entry in the core and complete interaction networks contains the following tab delimited information: Gene 1 Id, Ensembl gene ID for human interaction partner 1; Gene 1 description, alternative names for human Gene 1 (from Ensembl); Gene 2 Id, Ensembl gene ID for human interaction partner 2; Gene 2 description, alternative names for human Gene 2 (from Ensembl); Source Organism, the model organism protein interaction dataset that predicts this human protein interaction; Ortholog 1, model organism interaction partner 1 from the model organism protein interaction that predicts the human protein interaction; Ortholog 2, model organism interaction partner 2 from the model organism protein interaction that predicts the human protein interaction; and Ortholog score, a confidence score for the human protein interaction based on the likelihood that the two human proteins are the functional orthologs of the two model organism proteins. The score ranges from 0 (no confidence) to 4 (high confidence). The score is calculated as the sum of the InParanoid confidence scores for each gene orthology assignment. A score of 4 means that both of the human genes and both of the model organism genes are all the main orthologs in their groups of co-orthologs according to InParanoid. These represent higher confidence human protein interactions. Description, this field contains the original annotation for the model organism protein interaction; for worm interactions this indicates whether the interaction is in the core dataset of interactions found more than once (CORE\_1), or interactions that reconfirmed when retested (CORE\_2), or non-core interactions that did not reconfirm (NON\_CORE) \[[@B6]\]. For fly interactions this indicates the interaction score. This score mainly depends upon the number of times each interaction was detected, the total number of interactions made by each protein and the local network clustering, see \[[@B5]\] for details. A score \>0.5 is considered high confidence. For yeast protein interactions, these are the annotations of von Mering et al. \[[@B11]\] and contain the following information: experimental/computation method (and the number of times the interaction was detected); Von Mering et al.\'s confidence assignment; and whether the interaction was previously known in the literature. For more information, please see \[[@B11]\]. Additional data file [3](#s3){ref-type="supplementary-material"} lists the accuracy and coverage of GO term predictions at different levels in the GO hierarchy; Additional data file 4 lists gene function predictions for 85 human genes of unknown function; Additional data file 5 lists human disease genes with predicted protein interactions; and Additional data file 6 lists gene function predictions for 55 human disease genes. Supplementary Material ====================== ::: {.caption} ###### Additional data file 1 A complete list of predicted human protein interactions ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 2 A list of all core human protein interactions ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 3 The accuracy and coverage of GO term predictions at different levels in the GO hierarchy ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 4 Gene function predictions for 85 human genes of unknown function ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 5 Human disease genes with predicted protein interactions ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 6 Gene function predictions for 55 human disease genes ::: ::: {.caption} ###### Click here for additional data file ::: Acknowledgements ================ We thank the Sanger Institute Web Team for construction of the web interface and Paul Kersey for providing a list of TrEMBL accessions for yeast proteins. B.L. is supported by a Sanger Institute Postdoctoral Fellowship and A.G.F is supported by the Wellcome Trust. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Sources of predicted human protein interactions. (a)The number of human protein interactions predicted by the interaction maps from each model organism. (b)The number of human protein interactions predicted by the core higher-confidence interactions from each organism. As explained in the text, core interactions are those that reconfirmed when retested (worm), or had an interaction score of greater than 0.5 (fly) or were identified more than once in a single assay (yeast, worm). ::: ![](gb-2004-5-9-r63-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Filtering interaction datasets to improve their accuracy. (a)The percentages of interactions sharing GO terms at various depths in the GO hierarchy are compared for interactions predicted by the high-confidence interactions from each model organism (core yeast, core worm and core fly), as well as for the complete datasets from each organism (all yeast, all worm, all fly). For comparison, the percentage of shared GO terms is shown for a randomly generated network of the same size as the complete human network (random pairs). The x-axis indicates the depth in the GO hierarchy being considered, and the y-value the percentage of interaction partners (with known GO annotations) that share GO annotations at this depth or deeper. (b)The percentages of interactions sharing GO terms at different levels in the GO hierarchy are compared for interactions predicted by core interactions in two or more species (two species (core)), by interactions in the complete datasets of two or more species (two species), for interactions predicted by more than one experimental method in yeast (two methods), by any core interaction (all core), by any interaction (all), or by a randomly generated interaction network of the same size as the complete human interaction network (random pairs). All values shown are the percentage of non-self interactions between pairs of proteins that both have at least one associated GO term at the indicated depth in the GO hierarchy. ::: ![](gb-2004-5-9-r63-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The number and accuracy of human protein interactions predicted by different model organism protein-interaction datasets ::: Data source Predicted human interactions Interactions sharing GO terms -------------------- ------------------------------ ------------------------------- ------ All 71,496 12,724 24.9 Yeast 55,231 10,727 26.2 Fly 12,059 1,404 19.0 Worm 4,494 753 24.4 All core 11,487 3,133 38.1 Core yeast 6,061 2,146 45.4 Core fly 2,889 488 27.8 Core worm 2,701 597 32.3 Two species 288 154 74.8 Two species (core) 160 95 88.0 Two methods 2,166 829 60.6 Random pairs 71,496 6,053 14.6 The table lists the total number of interactions predicted by each interaction dataset, and the number of these interactions that connect proteins that share at least one GO term (at level 3 or deeper in the GO hierarchy). The percentages are relative to the total number of non-self interactions where both proteins have at least one GO annotation. All, all predicted human protein interactions; Yeast/worm/fly, interactions predicted by the yeast, worm or fly interaction maps; All core, all interactions predicted by the high-confidence subsets of each model organism interaction map (see Materials and methods); Two species, interactions predicted by more than one model organism interaction map; Two species (core), interactions predicted by the high-confidence subset of interactions from more than one model organism; Two methods, interactions predicted by data derived from more than one different interaction assay; Random pairs, the data for a randomly generated interaction network. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### The number of interactions, genes, novel genes and disease genes in the complete and core human interaction networks ::: Network Interactions Genes Novel genes Disease genes ---------- -------------- ------- ------------- --------------- Complete 71,496 6,231 1,482 448 Core 11,487 3,872 864 292 The complete network consists of all human protein interactions predicted by model organism protein-interaction datasets. The core network consists of all the human interactions predicted by the high-confidence subsets of each interaction network (see Materials and methods). Novel genes are defined as those without GO annotations. Disease genes are defined by the OMIM database \[25\], available from Ensembl \[16\]. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The approximate accuracy and coverage of GO terms predicted by the core and complete interaction networks ::: Number of interactors with GO term Core data Complete data ------------------------------------ ----------- --------------- ---- ---- 1+ 8 26 3 35 2+ 22 11 8 19 3+ 30 7 11 14 4+ 36 5 15 11 5+ 42 4 18 8 6+ 45 3 20 7 The approximate accuracy and coverage of GO term predictions were calculated for every gene in the core or complete interaction networks with at least one known GO term. The GO terms of a gene are predicted using the GO terms of any of its interaction partners (1+), or GO terms shared by at least two to six of its interaction partners (2+ to 6+). Accuracy is calculated as the number of correctly predicted GO terms divided by the total number of predicted GO terms. Coverage is calculated as the number of correctly predicted GO terms divided by the total number of known GO terms associated with each gene. These values are similar for GO annotations at different levels of the GO hierarchy (see Additional data file 3). ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Sources of model organism protein-interaction data ::: Dataset Interactions Type Reference URL --------- ----------------- ------------------------------ ----------- -------- Fly 20,020 Two-hybrid \[5\] \[26\] Worm 4,605 Two-hybrid \[6\] \[27\] Yeast 78,391 Total \[11\] \[11\] 5,125 Two-hybrid 49,313 Complex purification 886 Genetic 23,844 (23,399) In silico(In silicoonly) The table lists the total number of interactions contained in each model organism dataset, together with the method used to identify interactions, the publication reference, and the website (URL) from which the interaction dataset was obtained. For each dataset, the non-redundant number of unique interactions between unambiguously identified proteins is shown. For the yeast interactions, the total number of interactions is shown, as well as the number of interactions identified using each detection method. In silicoonly are interactions only predicted by in silicomethods without any confirmation from the experimental datasets. :::	PubMed Central	2024-06-05T03:55:48.493027	2004-8-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522870/", "journal": "Genome Biol. 2004 Aug 13; 5(9):R63", "authors": [ { "first": "Ben", "last": "Lehner" }, { "first": "Andrew G", "last": "Fraser" } ] }
PMC522871	Background ========== Genes that have recently fallen out of use for an organism are often detectable in the genome as pseudogenes - disabled copies of genes characterizable by disruptions of their reading frames due to frameshifts and premature stop codons \[[@B1]-[@B3]\]. Surveys of the pseudogene populations of eukaryotes (budding yeast, nematode worm, fruit fly and human) have recently been completed \[[@B4]-[@B10]\]. These pseudogene analyses have yielded insights into eukaryotic proteome evolution, showing that duplicated pseudogene formation tends to occur in younger, more lineage-specific, protein families, and is in many cases linked to the generation of functional diversity \[[@B3]\]. However, pseudogene formation in most prokaryotes has not been analyzed as a matter of course, and has, historically, been assumed to be minimal \[[@B11]\]. Some recent substantial populations of pseudogenes have been discovered in pathogenic bacteria, most notably in the leprosy bacillus Mycobacterium leprae, where around 1,100 pseudogenes (compared to around 1,600 genes) were found, with pseudogene formation providing a \'fossil record\' of recent wholesale loss of pathways involved in lipid metabolism and anaerobic respiration \[[@B12]\]. Here we want to address the question of whether these large populations are exceptional, or whether there are substantial populations of pseudogenes in other prokaryotic genomes. If so, from a holistic \'polygenomic\' perspective, what sorts of proteins tend to form prokaryotic pseudogenes? And are there any themes in common with the occurrence of pseudogenes in eukaryotes? To address these broad questions, we have adapted a pipeline developed for eukaryotic pseudogene identification to 64 prokaryotic genomes \[[@B4]\]. The species analyzed include archaea, pathogenic bacteria and non-pathogenic bacteria, and many of the pathogenic bacteria are also important organisms in current biodefense research. We have found nearly 7,000 pseudogenes, with notable numbers of pseudogenes for specific families linked to DNA transposition and also that have some role in environmental responses. Our results, which we have derived consistently across all the genomes, are available from our prokaryote pseudogene information website \[[@B13]\]. Results and discussion ====================== Pseudogenes are pervasive in prokaryotes ---------------------------------------- To identify pseudogenes in prokaryotic genomes, we performed a conservative and comprehensive search, as outlined in Figure [1](#F1){ref-type="fig"} and Materials and methods. We used a proteome set consisting of sequences from the 64 genomes and Swiss-Prot \[[@B14]\] with relatively high confidence in annotation (that is, excluding those annotated as hypothetical proteins). Intergenic regions in prokaryotic genomes were searched against the proteome set using FastX \[[@B15]\] for homology matches with disablements as pseudogene candidates. We then applied several checks to reduce false positives (see Materials and methods). Overall, we found 6,895 candidate pseudogenes. Previously, the pseudogene fraction was defined as the ratio of the number of pseudogenes to the number of all gene-like sequences (genes plus pseudogenes) \[[@B16]\]. By this measure, we find that pseudogenes are pervasive in prokaryotes (Figure [2](#F2){ref-type="fig"}). Pseudogenes are detectable at a low \'background\' level in most prokaryotes, ranging from 1 to 5% of the genome (Figure [2](#F2){ref-type="fig"}). Application of a more restrictive cutoff (E-value less than 0.001, instead of E-value less than 0.01) in FastX alignment results in slightly smaller percentage of pseudogenes (0.1% less on average) in all the genomes, and generates essentially the same results (data not shown). Our census is in general agreement with previous assessments of pseudogene content in the genomes of M. leprae, Escherichia coliand Rickettsia prowazekii\[[@B12],[@B16]-[@B19]\]. In these previous studies, however, different criteria were used for pseudogene identification in different genomes, leading to inconsistencies in comparing results. This is avoided in our study by using a method applied uniformly across all genomes. All these assessments suggest that most prokaryotes have similar net genomic DNA deletion rates, resulting in similar low-level \'background\' pseudogene fractions in their genomes. To check for a correlation with microbial \'lifestyle\', we classified the 64 species into three categories: archaea, pathogenic bacteria and non-pathogenic bacteria. The pseudogene fractions for these groupings were assessed. M. lepraehas a very large pseudogene fraction (36.5%) and is clearly a unique outlier. When this genome is set aside, the three groups have similar pseudogene fractions (3.6%, 3.9% and 3.3%). Note that three other pathogenic species/strains have relatively large pseudogene fractions, including Neisseria meningitidisMC58 (12.4%), N. meningitidisZ2491 (11.6%) and Rickettsia conorii(9.7%). The higher pseudogene fractions of some pathogenic species have previously been suggested to be a result of a rapidly changing environmental niche, with loss of metabolic and respiratory pathways \[[@B12]\]. We found that about 2,300 of our 6,895 candidate pseudogenes overlap with more than 2,600 annotated hypothetical open reading frames (ORFs), whose fractions were indicated in Figure [2](#F2){ref-type="fig"}. The overlap could arise from erroneous gene annotations or sequencing errors \[[@B16]\]. In either case, the pseudogene annotation in prokaryotic genomes is evidently an important part of decontaminating gene annotation. Pseudogene families ------------------- We used the Pfam classification \[[@B20]\] to analyze the families and functions of candidate pseudogenes. The 20 top-ranking domain families in terms of pseudogenes are shown in Figure [3a](#F3){ref-type="fig"}. Many large divergent gene families are among the top pseudogene families, including 9 of the top 10 gene families such as: the ABC transporter (PF00005), short-chain dehydrogenases/reductases (PF00106), sugar transporter (major facilitator superfamily) (PF00083), and histidine kinase-like ATPase (PF02518). As the largest family of proteins in prokaryotes, the ABC transporter functions to translocate a variety of compounds across biological membranes \[[@B21]-[@B23]\]. It consists of two ATP-binding domains (PF00005) \[[@B24],[@B25]\] and two transmembrane domains (PF00664). These domains are present in large copy numbers across genomes (2,172 and 245 gene copies as well as 67 and 13 pseudogene copies respectively). There are notable protein families that rank high in pseudogene number, but low in terms of gene number. They include the PPE family (PF00823) which is thought to be linked to antigenic variation in mycobacteria and is highly polymorphic \[[@B26]\]; the cytochromes P450 (PF00067), which are involved in processing diverse substrates; the GGDEF domain (PF00990), which is of unknown function and is associated with a wide diversity of other protein domains \[[@B27]\]; alpha/beta-hydrolase enzymes (PF00561), which have diverse catalytic functions; and pseudo-U-synthase-2 enzymes (PF00849), which help synthesize pseudouridine from uracil. Note that the first two families in this list have sequence diversity that has some link to environmental response. Figure [3b](#F3){ref-type="fig"} shows the relationship between the number of pseudogenes and genes for Pfam families. One might expect this relationship to be linear, with bigger families having more pseudogenes, but Figure [3b](#F3){ref-type="fig"} shows this is not the case. Two large families that have a relatively high ratio of pseudogenes to genes are the transposase DDE domain (PF01609) and integrase core domain (PF00665). Transposase facilitates DNA transposition and horizontal gene transfer and its DDE domain may be responsible for DNA cleavage at a specific site followed by a strand-transfer reaction \[[@B28]\]. Many transposons contain transposases for their transposition \[[@B29],[@B30]\]. We found that two strains of N. meningitidis(MC58 and Z2491) carry 26 and 22 copies of transposase pseudogenes, respectively, and have only 11 and 5 copies of transposase genes. In the MC58 strain, transposase pseudogenes have been found in most of the 29 remnant insertion sequences \[[@B31]\]. This suggests that N. meningitidisstrains probably undergo high selection pressure for transposases. The integrase core domain family (PF00665) is the catalytic domain of integrase, which mediates integration of a DNA copy of a viral/bacteriophage genome into the host genome \[[@B32]\]. It catalyzes the DNA strand-transfer reaction by ligating the 3\' ends of the viral DNA to the 5\' ends of the integration site \[[@B32]\]. The large number of transposase and integrase pseudogenes might result from harmful foreign genes being disabled in transposable elements. Several species contain many integrase pseudogenes, including Streptococcus pneumoniae, M. leprae, M. tuberculosis, and E. colistrain O157:H7. The large number of pseudogenes relative to genes for these two gene families may reflect an overall high selective pressure for them - that is, a gene family that is rapidly duplicating and evolving may generate many pseudogenes. Origins of pseudogenes ---------------------- Retrotransposition and genomic DNA duplication generate pseudogenes in mammals and other eukaryotes \[[@B2],[@B3]\]. In contrast, in prokaryotes, based on the experience annotating E. coliand M. leprae\[[@B12],[@B16]\], pseudogenes are suggested to arise from three process: the disablement of detectable native duplications; the decay of native single-copy host genes; and failed horizontal transfers. However, the complete extent of the processes forming prokaryotic pseudogenes is not yet well understood. We realize that there are many methods of defining horizontal transfer \[[@B33]-[@B36]\] and an active debate on the best way of doing this \[[@B37],[@B38]\], so we applied two independent methods to predict horizontal gene transfer events. The first method (GC-content) is based on the GC content bias at particular codon positions of recently acquired genes \[[@B33],[@B39]\]. The second method (GeneTrace) is based on the analysis of phylogenetic distribution of protein families on species tree \[[@B40]\]. In the GC-content method, the number of pseudogenes resulting from horizontal transfer in each genome was estimated by applying the same criteria to them as had been previously used to identify horizontally transferred genes. Overall, we found that the ratio (19.9%) of pseudogenes from potential horizontal transfer to those derived from the host is significantly higher than the ratio of genes in the host (8.6%). We dubbed the ratio of these two quantities the \'failed horizontal transfer index\', and observed that it implies that pseudogenes are 2.3 times more likely to arise from horizontal transfer than host genes are (Table [1](#T1){ref-type="table"}). To confirm our findings based on a method relying on GC content bias we applied the GeneTrace method (see Materials and methods). We analyzed a subset of pseudogenes and found that 18% result from failed horizontal transfer events, consistent with the previous method. Note that GeneTrace and the GC-content method are very different in the criteria they use to assess horizontal transfer and thus make for good independent verification of each other. In summary, we report here for the first time an estimate of how often horizontal transfer in prokaryotes introduces genes that are redundant, useless or even detrimental. Firstly, ORFs from dangerous genetic elements are under strong selection pressure to be deleted from the host\'s genome \[[@B11]\]. Secondly, horizontally transferred genes have a higher chance than non-transferred genes of becoming pseudogenes in most prokaryotes, which may be a result of deactivation/disablement of non-beneficial transferred genes. By examining closely related strains of the same species, we found that most close strains have a similar value for the failed horizontal transfer index. In particular, M. tuberculosis(strains H37Rv and CDC1551), N. meningitidis(strains Z1491 and MC8), and Helicobacter pylori(strains 26695 and J99) share similar index values within species. However, E. colihas different index values in the three strains studied. The free-living E. coliK12 strain has an index value of 4.6, comparable to values calculated from previous results \[[@B16]\], while the two pathogenic E. colistrains O157:H7 and O157:H7 EDL933 have much lower values (1.8 and 0.8). This can be readily explained in two ways: the intracellular pathogenic E. colistrains could have moved into a different environment that results in lower exposure to incoming DNA and thus to a lower rate of horizontal gene transfer \[[@B41]\]; or these strains could have an increased rate of gene loss or pseudogene formation of their host genes. A polygenomic power-law-like trend in pseudogene disablement ------------------------------------------------------------ To characterize the overall rate of decay of pseudogene populations, we plotted the fraction of disablements versus the average number of matching residues (to their closest homologs) per pseudogene for each species. This measure shows how the overall level of decay of a pseudogene population relates to age (which corresponds to the degree of overall match to the closest homologs). There is a general power-law-like behavior governing this measure, with recent pseudogenes having few disablements and divergent pseudogenes having many (Figure [4](#F4){ref-type="fig"}). Archaea and most non-pathogenic bacteria cluster together at higher rates of disablement (between 10 and 28 per 1,000 residues) and less significant matches, indicating comparatively greater retention of ancient gene remnants in those species and fewer young pseudogenes. On the other hand, obligate pathogenic bacteria tend to have younger pools of pseudogenes, even though they exhibit high disablement rates. Interestingly, four species of obligate bacterial pathogens clearly stand out from the general tendency: these are M. lepraeand three closely related mycoplasma species: Mycoplasma pneumoniae, Mycoplasma pulmonisand Ureaplasma urealyticum. Pseudogenes in these four pathogenic bacteria carry several times more disablements, suggesting that these bacteria have an accelerated disabling mutation rate. It is known that M. lepraehas lost the dnaQ-mediated proofreading activities of DNA polymerase III \[[@B12],[@B42]\], which could contribute to a higher mutation rate. The higher mutation rates in these species might suggest that these pathogens are under adaptation to their new environment, or have specific genome regions that are hypermutable. It is important to note here that the current sequence databases are derived from an uneven sampling of genomes. Therefore, genomes of organisms with more sequenced relatives may appear to have, on average, a seemingly younger population of pseudogenes, while others may appear to have older and fewer identifiable pseudogenes. Using data from 64 genomes, our results indicate an overall trend for pseudogenes observed in most of the genomes studied. However, these results have to be viewed as preliminary until more genome data is available. Conclusions =========== We have shown that pseudogenes in prokaryotes are not uncommon, occupying 1-5% of all gene-like sequences. We find that specific gene families with clear links to DNA transposition and environmental responses have higher pseudogene/gene ratios. The pseudogene data has many implications for the study of genome reduction and expansion \[[@B43],[@B44]\]. A significant proportion of the pseudogenes arose from putative failed horizontal transfer - at more than two times the rate for genes. Obligate pathogenic bacteria have high rates of disablement in younger pseudogene populations, consistent with recent accelerated genome reduction \[[@B44]\], while, in contrast, archaea and non-pathogenic bacteria have relatively older pseudogene populations, but similar rates of disablement. In terms of methodological implications, it is evidently necessary to include prokaryote pseudogenes as part of systematic annotation pipelines in the future. In addition, it was also shown to be helpful to identify potential short ORFs \[[@B45]\]. Furthermore, our survey shows that trends can be observed \'polygenomically\' for prokaryotes, where they are not obvious or significant in individual genomes. Materials and methods ===================== Database releases used ---------------------- We used the following datasets in our prokaryotic pseudogene analysis: Swiss-Prot (release 40.19 and updated to 27 May, 2002) \[[@B14]\] containing 43,094 prokaryotic protein sequences; nucleotide sequences from 64 prokaryotic genomes from EMBL database release 70 on March-2002 \[[@B46]\], including 11 genomes from archaea and 53 from bacteria as listed in Figure [1](#F1){ref-type="fig"}; Pfam release 7.3 of May 2002, containing 3,849 families and 498,152 protein domains in the alignments \[[@B20]\]. Pseudogene identification pipeline ---------------------------------- Figure [1a](#F1){ref-type="fig"} shows the basic procedure for identifying prokaryotic pseudogenes. The general schema was adapted from pipelines for pseudogene analysis in eukaryotes \[[@B4]\]. We generated a prokaryotic proteome set by collecting all the prokaryotic protein sequences in the Swiss-Prot database and those annotated in the 64 prokaryotic genomes. To be conservative, we did not include hypothetical or putative proteins, a large proportion of which might be overannotated \[[@B47],[@B48]\]. All the protein sequences were masked by SEG using the default low-complexity filter parameters (122.22.5) \[[@B49]\]. To maximize the efficiency of the pseudogene search, we only considered the intergenic DNA regions in the 64 prokaryote genomes (including the regions encoding hypothetical proteins) as query sequences, and searched their forward and reverse complement sequences against the proteome set using FastX \[[@B15]\]. Significant homology matches (E-value less than 0.01) that contained more than one disablement (either a frameshift caused by insertion or deletion of nucleotides or a premature stop codon) were considered as potential pseudogenes. If an intergenic region had multiple matches, these matches were sorted by E-value (increasing) and then by the number of matching residues (decreasing), if they have the same E-value. The match with the most significant E-value and the maximum matching residues was selected and redundant matches were removed. To ensure that spurious disablements were not introduced at ends of sequences as an alignment artifact, we excluded homology matches whose disablements occurred only within a \'cutoff region\' at either end. We used 16 residues for the cutoff region for short sequences (160 amino acids or fewer) - a parameter that has been applied previously \[[@B6]\]. For longer sequences (more than 160 amino acids), 10% of the sequence length was applied as the cutoff region as FastX tends to include more residues at the ends of alignments. We also assessed the potential pseudogenes by examining the distribution of the disablements within pseudogene sequences. Given that mutations within pseudogenes are unconstrained, we would expect disablements on pseudogenes to be evenly distributed. Figure [1b](#F1){ref-type="fig"} shows the position of disablements within pseudogene fragments whose length is normalized to 100 residues. By removing those potential pseudogenes that only had disablements at their flanking regions at both ends, the distribution is almost evenly distributed. We used it as a \'control filter\' to minimize false-positive pseudogenes. In the final pseudogene set, the length of pseudogenes ranges from 33 to 4,969 amino acids, with a median length of 130 amino acids, as compared with the proteome set, where the length ranges from 7 to 10,920 amino acids with a median length of 291 amino acids. We considered non-standard codon usage in some bacteria, such as when TGA encodes tryptophan rather than a stop codon in mycoplasma species, including Mycoplasma pneumoniae, M. pulmonisand U. urealyticum. By manual examination of E. coligenes with translational frameshifts in the RECODE database \[[@B50]\], we found that those genes were included in coding sequences (CDS) and therefore were excluded from our pseudogene search. Sequencing errors could also be a potential problem in the detection of pseudogenes. However, this effect is expected to be small, as comparison of independently sequenced isolates of the same E. colistrains indicated that only about 7% of candidate pseudogenes could be due to sequencing error \[[@B16]\]. To further consider the possibility of sequencing error, we examined the stop codons in the pseudogenes detected in the S. pneumoniaegenome (frameshift positions are not considered as they are difficult to locate.). This genome and eight others found in the trace archive of the National Center for Biotechnology Information (NCBI) \[[@B51]\] and Ensembl \[[@B52]\] were all sequenced by TIGR. We selected S. pneumoniaeas a case study as it is a relatively big genome available in the archive. By adapting a previous method \[[@B53]\], we examined the overall quality values (Q) for each nucleic acid of stop codons in the pseudogenes. Pseudogene sequences were aligned to the archived sequences (≥ 95% identity), and the quality values for nucleotides in stop codons were summed up. We chose 10^-2^as a cutoff of the error rate (err = 10^SUM(-0.1Q)^) for all nucleic acids. The stop codons with all three nucleic acids above the cutoff were validated. Out of 116 pseudogenes in this genome, 73 were found to contain 150 stop codons in total. Using the available data in the trace archive, we identified 54 pseudogenes with stop codons being aligned with the original sequences, and validated 47 of these (87%). In addition, a similar fraction of stop codons (101 out of 116) was confirmed. Family classification of genes and pseudogenes ---------------------------------------------- All genes in the 64 genomes were assigned to Pfam families by cross-referencing of their Swiss-Prot ID. Pseudogenes were assigned to Pfam families through ID of their closest homologs. Only the homologs that cover more than 70% of the Pfam domain were selected. A pseudogene could be assigned to multiple Pfam families if it contains multiple domains. Estimation of horizontally transferred genes and pseudogenes ------------------------------------------------------------ Here we used a method (GC-content) to estimate horizontal transferred genes on the basis of their base compositions \[[@B33],[@B39]\]. We analyzed each of the 64 genomes individually, and atypical genes and pseudogenes were identified if the GC content at first and third codon positions was two or more standard deviations higher or lower than the mean values at those positions in genes. To ensure that we had the codon positions accurately assigned for the GC-content method, we only analyzed codons for pseudogenes that aligned well with annotated protein sequences, specifically excluding the regions of the alignment around frameshifts. While it is true that the local alignment in some regions of a pseudogene may be ambiguous, causing some difference in the GC-content calculation in that region, the impact on the overall GC-content estimation is minimal, given how many positions we average over to calculate the failed transfer index score. The results for the 64 genomes are shown in Table [1](#T1){ref-type="table"}. The failed transferred index in the last column represents the ratio of the fraction of putative horizontally transferred pseudogenes to the fraction of horizontally transferred genes ![](gb-2004-5-9-r64-i1.gif), similar to the measure previously used in E. coli\[[@B16]\]. This essentially gives a likelihood ratio for horizontal transfer for pseudogenes relative to that of genes. Note that to minimize the effect of more divergent sequence alignments, for the horizontal-transfer calculations we only analyzed 1,748 \'recent\' pseudogenes, which have more than 50% sequence identity to their closest matches over an aligned subsequence of more than 100 residues. We have investigated the statistical robustness of the failed transfer index using resampling approaches \[[@B54]\]. For each of the 64 genomes, we randomly picked 90% of its genes and calculated their GC content. Using the new GC content, we then identified the putative horizontally transferred genes and pseudogenes and calculated the failed transfer index. We applied the process 1,000 times, generating a distribution of 1,000 indexes, which has a mean value of 2.32 with standard deviation of 0.01. We also applied an alternative method (GeneTrace) to estimate horizontally transferred pseudogenes \[[@B40]\]. In this method, potential horizontal transfer events are inferred within a protein family when it is present only in distantly related species and is absent from members of the same phylogenetic clade. We analyzed a subset of pseudogenes - 225 pseudogenes across 62 genomes - whose closest Swiss-Prot homologs share more than 70% sequence identity across at least 100 amino acids, and identified 41 of them (18%) as from failed horizontal transfer events. Acknowledgements ================ M.G. thanks NIH/NIAID grant for Northeast Biodefense Center (1U54AI057158-01) for financial support. He also acknowledges support from the Ruth B. Williams Fund. Y.L. was partially supported by an NLM postdoctoral fellowship (NIH Grant T15 LM07056). We thank Zhaolei Zhang and Nick Carriero for helpful discussions and Duncan Milburn for technical help. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Pseudogenes in prokaryotes. (a)Procedure for assigning pseudogenes. The flow chart shows the steps in identifying pseudogenes in 64 prokaryote genomes. The steps include: separate intergenic regions from coding sequence (hypothetical ORFs were excluded); six-frame FastX search on intergenic regions for pseudogene candidates; quality control to reduce false-positive results introduced by artificial disablement or by different codon usage. (b)The occurrence of relative disablement positions in pseudogenes, which were normalized on a 100-residue scale based on ratios of the distances from starting residues to disablements to the length of pseudogenes. The yellow bars indicate the distribution of disablement positions before the last quality-control step and the green bars show the distribution after minimizing false-positive pseudogenes. ::: ![](gb-2004-5-9-r64-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Fractions of pseudogenes in the 64 prokaryote genomes. The genomes are divided into three categories: archaea (green), non-pathogenic bacteria (blue) and pathogenic bacteria (purple). The yellow bars represent the fractions of pseudogenes that overlap with hypothetical ORFs, and the green bars represent those that do not overlap. Genomes in each category are sorted by the green bars. ::: ![](gb-2004-5-9-r64-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Gene-to-pseudogene ratios. (a)The top 20 pseudogene families and top 10 gene families based on Pfam classification. Ranking is based on the size of pseudogene families. The top 10 gene families are highlighted with the green background. (b)The number of genes plotted against the number of pseudogenes in a Pfam family. The line represents the overall ratio of the number of pseudogenes to the number of genes in the 64 genomes. ::: ![](gb-2004-5-9-r64-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The fraction of disabled residues (per 1,000 residues) versus the number of average matching residues to the closest homologs per pseudogene in the 64 species categorized into four groups: archaea (blue diamonds), non-pathogenic bacteria (green squares), obligate pathogenic bacteria (purple circles) and non-obligate pathogenic bacteria (red triangles). ::: ![](gb-2004-5-9-r64-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Putative horizontally transferred genes and pseudogenes ::: Species Gene Pseudogene Failed transfer index -------------------------------- --------- ------------ ----------------------- ----- ------ Archaea A. pernix 615 45 4 2 6.8 S. solfataricus 2,235 231 48 6 1.2 S. tokodaii 1,797 185 35 19 5.3 P. aerophilum 1,855 171 10 3 3.3 Halobacterium sp. NRC-1 1,383 100 1 1 13.8 M. thermautotrophicus 1,350 122 5 5 11.1 M. jannaschii 1,280 106 15 8 6.4 P. abyssi 891 75 6 2 4.0 P. horikoshii 553 50 8 0 0.0 T. acidophilum 1,169 106 5 4 8.8 T. volcanium 1,061 100 16 6 4.0 Non-pathogenic bacteria A. aeolicus 1,244 107 3 0 0.0 Synechocystis sp. PCC 6803 2,696 237 5 1 2.3 Nostoc sp. PCC 7120 3,672 332 10 2 2.2 S. coelicolor 6,012 536 14 4 3.2 B. halodurans 3,279 299 11 3 3.0 B. subtilis 1223 102 44 3 0.8 L. innocua 2,924 263 1 1 11.1 C. acetobutylicum 3,129 295 5 1 2.1 L. lactis subsp. lactis 1,870 156 13 2 1.8 C. vibrioides 2,699 231 6 1 1.9 M. loti 5,235 476 14 3 2.4 S. meliloti 2,985 240 9 6 8.3 E. coli K12 2,897 230 63 23 4.6 T. maritima 1,445 137 8 0 0.0 D. radiodurans 1,964 134 9 1 1.6 Pathogenic bacteria Buchnera sp. APS 477 42 5 2 4.5 U. urealyticum 467 40 2 1 5.8 M. pneumoniae 610 55 30 19 7.0 B. burgdorferi 590 63 1 0 0.0 M. pulmonis 595 53 2 1 5.6 C. trachomatis 597 67 3 1 3.0 C. muridarum 815 81 2 0 0.0 R. prowazekii 504 49 7 1 1.5 T. pallidum 727 64 12 5 4.7 C. pneumoniae J138 839 74 1 0 0.0 C. pneumoniae AR39 831 70 5 1 2.4 C. pneumoniae CWL029 845 71 7 0 0.0 R. conorii 695 67 9 0 0.0 M. leprae 1,440 119 271 53 2.4 C. jejuni 1,291 108 2 0 0.0 H. pylori J99 856 70 5 1 2.4 H. pylori 26695 1,055 90 13 3 2.7 S. pyogenes M1 GAS 1,348 108 14 1 0.9 S. pneumoniae 1,632 114 54 2 0.5 N. meningitidis Z2491 1,432 112 26 4 2.0 P. multocida 1,035 96 7 2 3.1 N. meningitidis MC58 1,466 121 44 14 3.9 X. fastidiosa 1,550 152 15 1 0.7 S. aureus subsp. aureus N315 1,557 140 4 2 5.6 S. aureus subsp. aureus Mu50 1,563 138 4 2 5.7 L. monocytogenes 2,799 231 2 0 0.0 C. perfringens 1,943 165 2 0 0.0 B. melitensis 2,948 216 5 0 0.0 R. solanacearum 3,032 252 5 0 0.0 V. cholerae 2,846 216 24 5 2.7 M. tuberculosis CDC1551 2,837 262 49 7 1.5 M. tuberculosis H37Rv 1,446 130 38 4 1.2 Y. pestis 3,533 282 51 4 1.0 S. typhi CT18 3,986 338 147 18 1.4 S. typhimurium LT2 4,308 349 22 5 2.8 E. coli O157:H7 3,424 266 120 16 1.7 E. coli O157:H7 EDL933 4,322 353 73 5 0.8 P. aeruginosa 3,716 281 7 3 5.7 Total 123,420 10,571 1,458 290 2.3 All genes and pseudogenes and the fraction having atypical codon-position-specific GC contents in the 64 genomes studied. The failed horizontal transfer index was computed as described in Materials and methods. :::	PubMed Central	2024-06-05T03:55:48.496292	2004-8-26	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522871/", "journal": "Genome Biol. 2004 Aug 26; 5(9):R64", "authors": [ { "first": "Yang", "last": "Liu" }, { "first": "Paul M", "last": "Harrison" }, { "first": "Victor", "last": "Kunin" }, { "first": "Mark", "last": "Gerstein" } ] }
PMC522872	Background ========== MicroRNAs (miRNAs) are non-coding RNA molecules with important regulatory functions in eukaryotic gene expression. The majority of known mature miRNAs are about 21-23 nucleotides long and have been found in a wide range of eukaryotes, from Arabidopsis thalianaand Caenorhabditis elegansto mouse and human (reviewed in \[[@B1]\]). Over 300 miRNAs have been identified in different organisms to date, primarily through cloning and sequencing of short RNA molecules \[[@B2]-[@B16]\]. Experimental miRNA identification is technically challenging and incomplete for the following reasons: miRNAs tend to have highly constrained tissue- and time-specific expression patterns; degradation products from mRNAs and other endogenous non-coding RNAs coexist with miRNAs and are sometimes dominant in small RNA molecule samples extracted from cells. Several groups have attempted to screen for new ArabidopsismiRNAs by sequencing small RNA molecules, but only 19 unique ArabidopsismiRNAs have been found so far \[[@B12],[@B13],[@B15]-[@B17]\]. While intensive research has unmasked several aspects of miRNA function, less is known about the regulation of miRNA transcription and precursor processing. A recent study shows a 116 base-pair (bp) temporal regulatory element located approximately 1,200 bases upstream of C. elegans let-7is sufficient for its specific expression at different developmental stages \[[@B18]\]. For some animal miRNAs, longer transcripts have been shown to exist in the nucleus before they are processed into shorter miRNA precursors \[[@B19]\]. Expressed sequence tag (EST) searches indicate that some human and mouse miRNAs are co-transcribed along with their upstream and downstream neighboring genes \[[@B20]\]. Most known animal miRNA precursors are approximately 70 nucleotides long, whereas the lengths of plant miRNA precursor vary widely, some extending up to 300 nucleotides \[[@B5],[@B8],[@B9],[@B14],[@B16]\]. As short mature miRNAs are generated from hairpin-structured precursors by an RNase III-like enzyme termed Dicer (reviewed in \[[@B21],[@B22]\]), evidence for miRNA expression based on the presence of longer precursor RNAs is likely to be found in genome-wide expression databases. Most known miRNAs are conserved in related species \[[@B5],[@B8],[@B9],[@B14]-[@B16]\]. Strong sequence conservation in the mature miRNA and long hairpin structures in miRNA precursors make genome-wide computational searches for miRNAs feasible. A variety of computational methods have been applied to several animal genomes, including Drosophila melanogaster, C. elegansand humans \[[@B4],[@B10],[@B11],[@B23]\]. In each case, a subset of computationally predicted miRNA genes was validated by northern blot hybridizations or PCR. A known function of miRNAs is to downregulate the translation of target mRNAs through base-pairing to the target mRNA \[[@B21],[@B24],[@B25]\]. In animals, miRNAs tend to bind to the 3\' untranslated region (3\' UTR) of their target transcripts to repress translation. The pairing between miRNAs and their target mRNAs usually includes short bulges and/or mismatches \[[@B26]-[@B28]\]. In contrast, in all known cases, plant miRNAs bind to the protein-coding region of their target mRNAs with three or fewer mismatches and induce target mRNA degradation \[[@B12],[@B15],[@B17],[@B29]\] or repress mRNA translation \[[@B30],[@B31]\]. Several groups have developed computational methods to predict miRNA targets in Arabidopsis, Drosophilaand humans \[[@B29],[@B32]-[@B35]\]. In the work reported here, we defined and applied a computational method to predict A. thalianamiRNAs and their target mRNAs. Focusing on sequences that are conserved in both A. thalianaand Oryza sativa(rice), we predicted 95 ArabidopsismiRNAs, including 12 of 19 known miRNAs and 83 new candidates. Northern blot hybridizations specific for 18 randomly selected miRNA candidates detected the expression of 12 miRNAs. The sequences of another eight predicted miRNAs were found in the public ArabidopsisSmall RNA Project (ASRP) database \[[@B36]\]. We also found massively parallel signature sequencing (MPSS) evidence for 14 known ArabidopsismiRNAs and 16 predicted ones. For 77 of the 83 predicted miRNAs we found putative target transcripts that were functionally conserved between Arabidopsisand O. sativa, with a signal-to-noise ratio of 4.1 to 1. Finally, we find supporting evidence for miRNA regulation of some mRNA targets using available genome-wide microarray data. The authentication of three predicted miRNA targets was validated by identification of the corresponding cleaved mRNA products. Results ======= Prediction of ArabidopsismiRNAs --------------------------------- To predict new miRNAs by computational methods, we defined sequence and structure properties that differentiate known ArabidopsismiRNA sequences from random genomic sequence, and used these properties as constraints to screen intergenic regions in the A. thalianagenome sequences for candidate miRNAs. Besides the well known hairpin secondary structure of miRNA precursors, the 19 unique known ArabidopsismiRNAs collected in Rfam \[[@B37]\] were evaluated for the following computable sequence properties: G+C content in mature miRNA sequences, hairpin-loop length in their precursor RNA structures, number and distribution of mismatches in the hairpin stem region containing the mature miRNA sequence, and phylogenetic conservation of mature miRNA sequences in the O. sativagenome. Sequences of all 19 known ArabidopsismiRNAs had a G+C content ranging from 38% to 70%. For 15 of the 19 miRNAs, the predicted secondary structure of their precursors, or at least one precursor if a miRNA has multiple genomic loci, had a hairpin-loop length ranging from 20 to 75 nucleotides. In the hairpin structures formed by miRNA precursors, all miRNAs were found in the stem region of the hairpin, and had at least 75% sequence complementarity to their counterparts. Fifteen of 19 miRNAs were conserved with at least 90% sequence identity in the O. sativagenome. Thus, constraints of G+C content between 38 and 70%, a loop length between 20 and 75 nucleotides, and a minimum of 90% sequence identity in O. sativawere used to predict ArabidopsismiRNA. The first step was to search for potential hairpin structures in the Arabidopsisintergenic sequences. As most known ArabidopsismiRNAs are around 21 nucleotides long, we used a 21-nucleotide query window to search each intergenic region for potential miRNA precursors as follows: for each successive 21-nucleotide query subsequence, if a 21-nucleotide pairing subsequence with more than 75% sequence complementarity was found downstream within a given distance (hairpin-loop length), the entire sequence from the beginning of the query subsequence to the end of the complement pairing subsequence with a 20-nucleotide extension at each side was extracted and marked as a possible hairpin sequence (see Materials and methods for details). The minimum and maximum hairpin-loop lengths used in this prediction were 20 and 75 nucleotides. Each 21-nucleotide query subsequence and its downstream complementary subsequence were considered as \'potential 21-mer miRNA candidates\' (referred to as \'21-mers\'). If a series of overlapping forward query sequences and their corresponding downstream pairing sequences were all identified from the same hairpin structure, each of them was initially considered as an individual 21-mer. The second step was to parse miRNA candidates according to their nucleotide composition and sequence conservation. A filter of G+C content between 38 and 70% was applied to all 21-mers obtained from the above step, followed by a requirement for more than 90% sequence identity in the O. sativagenome. The secondary structures of the resulting candidates were evaluated by mfold \[[@B38]\]. Only 21-mers whose Arabidopsisprecursor and corresponding rice ortholog precursor both had putative stem-loop structures as their lowest free energy form reported by mfold were retained. Because some non-coding RNA genes were not included in the current Arabidopsisgene annotation, orthologs of known non-coding RNA genes other than miRNAs were subsequently removed by aligning the 21-mers to non-coding RNAs collected in Rfam with BLASTN (version 2.2.6) \[[@B37]\]. The 21-mers that passed all sequence and structure filters above were considered as final miRNA candidates. A summary of the prediction algorithm is shown in Figure [1](#F1){ref-type="fig"}. In cases where two or more overlapping 21-mer miRNA candidates from the same precursor were collected in the final miRNA candidate set, each miRNA candidate was scored using the following formula: miRNA~score~= number of mismatches + (2 × number of nucleotides in terminal mismatches) + (number of nucleotides in internal bulges/number of internal bulges) + 1 if the miRNA sequence does not start with U. The term \'terminal mismatches\' in the formula above refers to consecutive mismatches among the beginning and/or ending nucleotides of a mature miRNA sequence. The term \'bulge\' refers to a series of mismatched nucleotides. Because the sequences of most known miRNAs start with a U, a U-start preference was used in the formula above by penalizing non-U-start sequences. The sequence with the lowest miRNA~score~from a series of overlapping 21-mers was selected as the final miRNA candidate. In total, we predicted 95 miRNA candidates in the Arabidopsisgenome, including 12 known ArabidopsismiRNAs and 83 new candidates. The former group corresponds to 63% of known ArabidopsismiRNAs to date (12 of 19). The remaining seven known miRNAs not included in the current prediction were filtered out as a result of their lower sequence conservation in the rice genome or longer loop length in their secondary structure, as outlined in Figure [1](#F1){ref-type="fig"}. Because of the complementarity between the two DNA strands of a given genome region, theoretically there should be two sequence possibilities for a predicted miRNA: the predicted sequence itself or, alternatively, its reverse complementary sequence located on the opposite strand of the genome. In many cases, however, owing to G::U pairing in RNA structure prediction, the complementary sequence of a miRNA precursor did not always exhibit a hairpin structure as its lowest energy folding form because the complement of a G::U pair, that is, C::A, altered the secondary structure. Thus, we were able to identify the coding strand of most predicted miRNA candidates through secondary structure evaluation. Furthermore, as described in the following sections, the sequences/partial sequences of some miRNA candidates or their precursors could be found in the ArabidopsisMPSS data used. As most MPSS data probably represent the expression of their associated miRNAs, we were able to use them to predict the miRNA coding strand. The coding strand of miRNA candidates that were contained in the ASRP database was determined according to cloned RNA sequences (see below for details). The complete list of predicted miRNAs is shown in Additional data file 1. Experimental validation of predicted miRNAs ------------------------------------------- To gain support for the expression of the predicted miRNAs, northern blot hybridizations were carried out using RNA samples from different tissues selected to cover a spectrum of potential miRNA expression patterns. Using strand-specific oligonucleotide probes, positive signals of expression were detected for 14 out of 18 miRNA candidates tested. The results for all newly identified miRNAs are shown in Figure [2a](#F2){ref-type="fig"} and [2b](#F2){ref-type="fig"}. Oligonucleotide probes against the antisense strand of different miRNA candidates were used as negative controls, and none produced any signal, as shown for miR417 in Figure [2b](#F2){ref-type="fig"}. Note that an extended exposure time was needed to detect expression of most miRNAs (indicated by a number in days in parentheses in Figure [2](#F2){ref-type="fig"}), suggesting that their abundance is significantly lower than that of other known miRNAs (that is, miR158 and miR159a in Figure [2c](#F2){ref-type="fig"}, and data not shown). In this analysis we also included 10 21-mers that were rejected by our miRNA prediction criteria as negative controls to evaluate the specificity of northern blot hybridization; as expected none of them produced a positive signal. The secondary structures of a few selected northern blot hybridization-positive miRNA candidates are shown in Figure [3](#F3){ref-type="fig"}. A full list of the secondary structures of predicted precursors of ArabidopsismiRNA candidates and their rice orthologs is available in Additional data file 2. Among the 14 miRNAs that produced positive signals in the northern blot hybridizations, two are close paralogs of known miRNAs; miR169b is a paralog of miR169 and miR171b is a paralog of miR170. Because it is impossible to distinguish closely related sequences by northern blot hybridization, we were unable to rule out the possibility that signals detected by probes for miR169b and miR171b were contributed by their known miRNA paralogs. However, as miR169b was also identified in the ASRP database (see next section), we were able to conclude that miR169b was a real miRNA. Thus, 12 candidates validated by northern blot hybridization should be annotated as bona fidemiRNAs (see Table [1](#T1){ref-type="table"} for a summary). Cloning evidence for predicted miRNAs ------------------------------------- An ASRP database has recently been made publicly available \[[@B36]\]. Sequences in the ASRP database were collected by cloning small RNA molecules with similar size to miRNAs and siRNAs \[[@B39]\]. To check whether any of our predicted miRNAs can be identified by a standard RNA cloning method, we compared the 83 predicted miRNA candidates with all sequences in the ASRP database. Eight newly predicted miRNA candidates were found in the ASRP database (Figure [4](#F4){ref-type="fig"}). Among them, five were identical to one or more cloned RNA molecules, indicating that we had correctly predicted the 5\' and 3\' ends and the actual length of these miRNA candidates. For the other three candidates, our predicted sequences were either shorter than, or a few nucleotides shifted from, their corresponding clones in the ASRP database. The exact sequences of these three miRNA candidates were then corrected according to the corresponding sequences in the ASRP database. The expression of miR169b and miR172b\* was also detected by northern blot hybridization (Figure [2a](#F2){ref-type="fig"}). Although miR169h was present in the ASRP database, it could not be detected by northern blot hybridization (see Additional data file 1). According to the current miRNA annotation criteria \[[@B22]\], these eight predicted miRNA candidates with corresponding cloned sequences in the ASRP database should be annotated as bona fidemiRNAs. MPSS evidence for known and predicted ArabidopsismiRNAs --------------------------------------------------------- To further validate the predicted miRNA molecules, we took advantage of available Arabidopsismassively parallel signature sequencing (MPSS) data. The MPSS sequencing technology identifies unique 17-nucleotide sequences present in cDNA molecules originated from polyadenylated RNA extracted from a cell sample. By inserting cDNA molecules into a cloning vector containing distinct 32-mer oligonucleotide tags, the MPSS technology ensures that each cDNA molecule is ligated to a unique tag and that more than 99% of the total cDNAs are represented after the cloning step. Tagged cDNAs are then amplified by PCR and hybridized to microbeads that have been precoated with multiple copies of unique anti-tags complementary to one type of 32-nucleotide tag. The expression level of a particular transcript is measured by counting the number of distinct microbeads that contain the same 17-nucleotide cDNA sequence. The MPSS technology does not require prior knowledge of a gene\'s sequence and thus can identify novel or rarely expressed genes. For a complete description, see \[[@B40],[@B41]\]. To assess the degree to which MPSS data could be used to support predicted miRNAs, we inspected the 19 known ArabidopsismiRNAs for unique representation in public ArabidopsisMPSS datasets and in our own MPSS datasets derived from a variety of tissues and conditions (see Materials and methods for details) \[[@B42]-[@B44]\]. We compared the intergenic genomic sequence flanking the 19 known ArabidopsismiRNAs with the MPSS data. We found 30 MPSS signature sequences that were identical to subsequences within the flanking 500-bp sequences either upstream or downstream of 14 known miRNAs (see Additional data file 3). All 30 MPSS sequences were reported in both the public and private MPSS datasets. They occurred upstream, downstream or partially overlapping with known mature miRNAs. Despite the highly repetitive nature of the Arabidopsisgenome, 28 of the 30 MPSS signatures mapped uniquely to only one miRNA locus, with no matches elsewhere in the genome. Two genomic loci were found for each of the two exceptional MPSS signatures MPSS78528 and MPSS28409. For MPSS78528, the associated miRNA mir162 appeared twice in the Arabidopsisgenome (upstream of At5g08180 and upstream of At5g23060) and the MPSS sequence mapped exactly to those regions. For MPSS28409, its second genomic match was on the opposite strand of an intron in gene At3g04740, which was very unlikely to be a source for MPSS sequences because samples for MPSS were prepared from mRNA or other type of polyadenylated RNA molecules, in which introns should have been processed. Thus, the MPSS data accurately reflected the expression of 14 known ArabidopsismiRNAs from a total of 19, indicating that it can be used as a source of indirect experimental support for the expression of predicted miRNAs. We then assessed the presence of MPSS signature sequences for the 83 predicted miRNAs. Using the approach described above, 23 MPSS signature sequences corresponding to the flanking sequences of 16 predicted miRNAs were found (see Additional data file 1). All 23 MPSS signature sequences were present in both the public and our own MPSS datasets, and mapped uniquely to the miRNA flanking sequence. The expression of nine miRNA candidates supported by MPSS data was also tested by northern blot hybridization, with eight of them producing a positive signal. Another three miRNAs with MPSS data were found in the ASRP database (see previous section and Additional data file 1). These results indicate that MPSS data indeed represent the expression of predicted miRNAs. Comparison of predicted miRNAs to known ArabidopsismiRNAs ----------------------------------------------------------- To explore the relationship of predicted miRNAs to known ArabidopsismiRNAs, we compared the sequences of all 83 miRNA candidates from our prediction with sequences of the 19 known ArabidopsismiRNAs. Eight predicted ArabidopsismiRNAs exhibited high sequence similarity to one or more known ArabidopsismiRNAs and could be grouped into five clusters (Figure [5](#F5){ref-type="fig"}). We could not find convincing evidence that Arabidopsisand animal miRNAs are related, as clustering of these required the insertion of multiple gaps in the alignments (data not shown). Putative mRNA targets of predicted ArabidopsismiRNAs ------------------------------------------------------ A previous study has predicted that most known plant miRNAs bind to the protein-coding region of their mRNA target with nearly perfect sequence complementarity, and degrade the target mRNA in a way similar to RNA interference (RNAi) \[[@B29]\]. Analysis of several targets has now confirmed this prediction, making it feasible to identify plant miRNA targets \[[@B12],[@B15],[@B16]\]. We developed a computational method based on the Smith-Waterman nucleotide-alignment algorithm to predict mRNA targets for the 83 newly identified miRNA candidates reported in this paper (see Materials and methods for details). Focusing on miRNA complementary sites that were conserved in both Arabidopsisand O. sativa, our method was able to identify 94% of previously confirmed or predicted mRNA targets for known conserved ArabidopsismiRNAs. Applying the method to the 83 predicted ArabidopsismiRNA candidates and their O. sativaorthologs, we predicted 371 conserved mRNA targets for 77 predicted ArabidopsismiRNAs, with an average of 4.8 targets per miRNA. The signal-to-noise ratio of the miRNA targets prediction was 4.1:1 when using randomly permuted sequences with the same nucleotide composition to miRNA sequences as negative controls that went through the same target prediction process. A complete list of these predicted target mRNAs and their pairings with miRNA sequences is available in Additional data file 4. Of the 371 predicted miRNA targets, 10 were potential targets of two independent miRNAs, one (At3g54460 mRNA) was a potential target of three different miRNAs (At1g60020\_5\_14, At3g27883\_1009, At5g62160\_613\_rc), and the rest were targets of a single miRNA. We assessed the biological functions of all predicted miRNA targets using gene ontology (GO) \[[@B45]\]. GO terms for 254 targets were found in the molecular function class. Molecular functions of the putative miRNA targets included transcription regulator activity, catalytic activity, nucleic acid binding, and so on, as summarized in Table [2](#T2){ref-type="table"}. As some proteins were classified in more than one molecular function category, the total number of targets listed in different function categories in Table [2](#T2){ref-type="table"} exceeds the number of targets with GO function assignment. Consistent with previous reports \[[@B29]\], a large proportion of predicted targets encoded proteins with transcription regulatory activity, corresponding to 50% of total targets with GO annotation (129/254). One interesting phenomenon was that most transcription regulators in the miRNA target set were plant specific, such as MYB, AP2, NAC, GRAS, SBP and WRKY family transcription factors (Table [3](#T3){ref-type="table"}). For example, the miRNA target set included 10 plant specific NAC-domain-containing transcription factors, corresponding to 9% of total NAC-domain-containing transcription factors encoded by the A. thalianagenome. In contrast, 139 genes encoding a general transcription factor bHLH were found in the A. thalianagenome, but only three were putative miRNA targets. We analyzed the expression patterns of potential targets to look for indications that they were under miRNA regulation. Twelve of the 14 miRNAs confirmed by northern blot hybridization showed an increased accumulation in flower tissue compared to the other tissues tested (Figure [2](#F2){ref-type="fig"}), suggesting a role for miRNAs in regulating flower-specific events. In a search of Arabidopsismicroarray gene expression data available from The ArabidopsisInformation Resource (TAIR) \[[@B46]\], we found the expression profile for 11 predicted mRNA targets that can base-pair nearly perfectly with five confirmed flower-abundant miRNAs. We hypothesized that expression levels of these targets in flower tissue could be decreased as compared to whole plant RNA samples as a result of mRNA cleavage induced by miRNA regulation. Accordingly, a reduced expression level (more than 1.25-fold decrement) was found for eight genes in total flower mRNA compared to total whole plant mRNA, with another three whose expression was almost unchanged (Table [4](#T4){ref-type="table"}). A t-test on the possibility of decreased expression between transcripts listed in Table [4](#T4){ref-type="table"} and in the entire microarray data resulted in a p-value of 0.04, indicating that the decreased expression observed for predicted miRNA targets is significantly different from the general expression pattern of the entire microarray data. Target mRNA fragments resulting from miRNA-guided cleavage are characterized by having a 5\' phosphate group, and cleavage occurs near the middle of the base-pairing interaction region with the miRNA molecule. Using a modified RNA ligase-mediated 5\' rapid amplification of cDNA ends (5\' RACE) protocol, we were able to detect and clone the At3g26810 mRNA fragment corresponding precisely to the predicted product of miRNA processing (Figure [6](#F6){ref-type="fig"}). Two other genes, At3g62980 (TIR1) and At1g12820, share extensive sequence homology with At3g26810 and were also predicted to be targets of miR393a. Consistent with this, we also identified the corresponding RNA fragments derived from miRNA cleavage by 5\' RACE (data not shown). We were not able to identify other targets from flower RNA samples using a similar approach. The microarray data used in this tissue comparison experiment includes around 7,400 genes only (about a quarter of the entire Arabidopsisgenome). Thus, we expect the expression profile of more mRNA targets to be determined as more whole-genome tissue comparison data is available. Discussion ========== We have developed and applied a computational method to predict 95 ArabidopsismiRNAs, which include 12 known ones and 83 new sequences. All 83 new miRNAs are conserved with more than 90% identity across the Arabidopsisand rice genomes. The expression of 19 new miRNAs was confirmed by northern blot hybridization or found in a publicly available database of small RNA sequences. MPSS data support was also found for 14 known and 16 predicted ArabidopsismiRNAs. Of the 16 miRNAs, 10 were confirmed by northern blot hybridization or by their presence in the ASRP database, and six have MPSS data only. In total, we have found direct or indirect experimental evidence for 25 predicted miRNAs. We expect more evidence to be found for other predicted miRNAs as independent experimental data, such as small RNA sequencing and MPSS data, grow. Among the 83 predicted miRNAs, eight have strong sequence similarity with known plant miRNAs. The prediction results and supporting experimental evidence are summarized in Table [5](#T5){ref-type="table"}. Additional data file 1 summarizes the corresponding evidence for known miRNAs and contains additional detailed information for each new candidate. Potential functionally conserved mRNA targets were found for 77 predicted miRNAs. Assessment of miRNA prediction ------------------------------ The prediction method developed in this study uses computable sequence and structure properties that characterize the majority of the known ArabidopsismiRNA genes to constrain the miRNA search space. Parameters used in the prediction were selected to minimize false positives while maximizing true positives. Thus, seven known miRNAs (37%) were missed using our selected parameters. However, relaxing the loop length range to include all known miRNAs increased the number of candidate hairpins from around 180,000 to around 337,000 (a 53% increase). As the method requires stringent miRNA sequence conservation between Arabidopsisand O. sativa, miRNAs with little or no sequence conservation in other genomes will be overlooked by this method. Given the current knowledge of miRNAs, it is difficult to develop computational methods to distinguish non-conserved miRNAs from noise. The prediction method developed here is not specific for Arabidopsisand can be applied to other pairs of related genomes as well. We attained a 67% success rate of northern blot hybridization on all tested miRNA candidates, demonstrating the expression of 12 miRNAs from a total of 18 tested candidates. Failure to detect miRNA candidates by northern blot hybridization could be due to the limited number of sample tissues tested, as specific miRNAs may be expressed only under particular conditions (stimuli and/or developmental stages) or in specific cell types. For instance, further analysis of miR169g\* (shown in Figure [2a](#F2){ref-type="fig"}) indicated a higher accumulation in mature siliques than in the seedling stage (J.L.R. and N-H.C., unpublished work). This can only be determined by detailed study of individual miRNAs, which is not possible in a general screening such as the one presented here. Alternatively, the expression level of certain miRNAs may fall below the detection limit of our assay conditions. Consistent with this idea, in all cases the confirmed miRNAs were detected only after an extended period of autoradiography (2-3 days), as compared with known miRNAs that were more easily detected (see Figure [2](#F2){ref-type="fig"} and data not shown). The low abundance of the newly identified miRNAs is in agreement with the limited number of miRNAs identified to date using the established cloning strategies. MPSS support for miRNAs and possible polyadenylation of precursor transcripts ----------------------------------------------------------------------------- We made an intriguing discovery by identifying known ArabidopsismiRNAs and their approximate precursor sequences in MPSS polyadenylated transcript datasets. All but two MPSS sequences reported here uniquely map to the mature miRNA sequence or to the flanking sequence 500 nucleotides upstream or downstream (see Results). The MPSS sequence locations and orientations indicate that they are not transcripts derived from surrounding annotated genes. All but one tested miRNA candidates with MPSS evidence produced positive signals in northern blot hybridizations. As MPSS cDNA libraries are generated using polyadenylated RNA molecules \[[@B40],[@B41],[@B43],[@B44]\], the presence of 14 known ArabidopsismiRNAs from a total of 19 in these datasets strongly indicates that at least some, if not all, plant miRNAs have a polyadenylated precursor form at some stage of their biogenesis. Predicted miRNAs detected by both northern blot hybridization and MPSS have consistent tissue-specific expression profiles under both methods. This supports the notion that the MPSS data reflect miRNA expression patterns. The public MPSS datasets are accessible only through an online interface that allows direct query of 17-nucleotide MPSS sequences. Direct comparison of the public sequences and predicted miRNAs was not possible. Thus, we were limited in our analysis to inquire whether a private MPSS sequence was also in the public MPSS dataset. Consequently, only MPSS sequences that appeared in the private set alone or in both sets were available to support miRNA prediction. The public MPSS dataset has 120-fold more signature sequences (more than 12 million additional tags) than our private MPSS dataset (approximately 94,000 tags total). Thus, we expect far more MPSS evidence for expression of the predicted miRNAs to be found in the public MPSS datasets when the public MPSS data are available to be compared locally. Target mRNAs for predicted miRNAs --------------------------------- We used phylogenetic conservation as a constraint on miRNA target mRNA prediction: only transcripts that had at least one O. sativafunctional ortholog among the top 500 rice miRNA targets were considered as potential miRNA target genes. Because all miRNA candidates reported here are highly conserved in rice, it is expected that their mRNA targets should also be conserved. Transcripts encoding proteins with ambiguous annotations, such as those for hypothetical proteins, expressed proteins and putative proteins, are not included in the target prediction because of the difficulty in identifying their orthologs in the O. sativagenome. Thus, the absence of predicted mRNA targets for the minority of the miRNAs may be due to the unfinished annotation of the O. sativagenome or to the divergence of target mRNA sequences that may preclude its identification. Gaps and mismatches are commonly seen in known animal miRNA::mRNA base-pairing interactions and, as a result, miRNA binding represses the translation of their targets \[[@B1]\]. Although in most known cases plant miRNAs tend to pair nearly perfectly with their target mRNAs and induce mRNA cleavage \[[@B12],[@B15],[@B17],[@B29]\], recent evidence has shown that plant miRNAs can also repress target mRNA translation in a way similar to that of animal miRNAs \[[@B1],[@B30],[@B31]\]. To further explore the function of ArabidopsismiRNAs in target mRNA translation repression, in this prediction we allowed gaps and mismatches in the putative ArabidopsismiRNA::mRNA pairs. The free energy of 90% of the predicted miRNA::mRNA pairs is lower than the average free energy of known animal miRNA::mRNA pairs (ΔG = -14 kcal/mol) \[[@B32]\], indicating that the predicted miRNA::mRNA pairs are potentially stable at the energy level if similar interactions are present in plants (see Additional data file 4). Our prediction of target mRNAs for the new 83 miRNA candidates reveals that a broad functional range of genes may be regulated by miRNAs (Table [4](#T4){ref-type="table"} and Additional data file 4). As in previous findings, the predicted miRNA targets were enriched for transcription factors \[[@B29]\]. Our prediction also included transcripts encoding proteins with transcription regulator activity, catalytic activity, signal transducer activity and translation regulator activity. This result is consistent with recent findings in animal miRNA targets, and suggests a broader role for miRNA regulation in plant gene expression \[[@B29],[@B32]-[@B35]\]. We took advantage of available microarray data to assess the relative expression levels of potential mRNA targets in tissues in which their miRNAs were expressed. For mRNA targets of several newly identified miRNAs, we found reduced expression levels in RNA samples from flowers compared to the whole plant. Accordingly, we identified the cleavage product of At3g26810 mRNA, and those of another two homologous genes, At1g12820 and TIR1 (At3g62980), to confirm them as targets of the same miRNA (miR393a). These genes encode F-box proteins, and TIR1 in particular is involved in auxin-mediated protein degradation \[[@B47]\]. Interestingly, F-box proteins are another group over-represented among the target mRNAs (48 out of 254, or 19%), while there are around 700 F-box proteins encoded in the Arabidopsisgenome (2.1%) \[[@B48]\]. Remarkably, these are the first confirmed plant miRNA targets that are not transcription factors, with the exception of DCL1 and AGO1. The identity and expression pattern of a target mRNA can help identify the specific expression profile of its corresponding miRNA. Tissues with low mRNA expression levels should be checked carefully for miRNA expression. Currently, this kind of search is limited by the availability of genome-wide and tissue-/time-specific microarray data. As such data accumulate, their analysis will enrich our understanding of the different biological processes regulated by microRNAs. Materials and methods ===================== Computational prediction of ArabidopsismiRNAs ----------------------------------------------- The Arabidopsisgenome version 3 and the O. sativagenome released by The Institute for Genome Research (TIGR) on 22 July 2002 \[[@B49]\] and 24 January 2003 \[[@B50]\], respectively, were used for the present study. Intergenic regions of both the A. thalianaand the O. sativagenomes were extracted according to the annotations provided by TIGR. A scanning algorithm implemented in the Perl programming language was used to search for possible hairpin structures within each intergenic sequence in A. thaliana. The method inspects each successive 21-nucleotide query window in each intergenic region, with two nucleotide increments, and searches for downstream complementary sequences with up to 25% mismatches. We restricted the distance from the last base of the forward query sequence and the first base of the downstream complementary pairing sequence to a minimum of 10 nucleotides and a maximum of 150 nucleotides (loop length). For each 21-nucleotide query, the loop length was increased one nucleotide at a time, and all downstream 21-nucleotide pairing sequences with more than 75% identity to the complement of the query sequence were considered as possible \'pairing sequences\'. For each query sequence, the downstream complementary sequence with the fewest mismatches was saved as the pair sequence for the query. Insertions and deletions were allowed in the alignment and were counted as mismatches. Sequences from the start of a qualified querying sequence to the end of its downstream complementary pairing sequence were considered as a potential miRNA precursor with putative hairpin structure. Twenty extra nucleotides were extracted from the genome at each end of a potential miRNA precursor for the purpose of structure check using mfold \[[@B38]\]. A G+C-content filter and a loop-length filter were applied to the 312,236 hairpin structures obtained. Only hairpins with a loop length between 20 and 75 nucleotides and 21-mer sequences with a G+C-content between 38 and 70% were analyzed further. The remaining 21-mer sequence pairs were aligned with rice intergenic regions using BLASTN (version 2.2.6) to identify homologous sequences in O. sativaintergenic regions with 90% or higher sequence identity. The secondary structure of ArabidopsismiRNA candidate precursors and their rice precursor orthologs was evaluated using mfold \[[@B38]\]. Only 21-mers whose Arabidopsisprecursor and rice ortholog precursor both had a hairpin-like folding as their lowest energy states were considered as miRNA candidates. A sequence alignment search against non-coding RNAs collected in Rfam \[[@B51]\] using BLASTN (version 2.2.6) was applied to identify and remove homologs of non-coding RNAs other than miRNAs. The remaining 95 sequences were retained as our final miRNA dataset. Northern blot hybridizations ---------------------------- Two-day-old seedlings, 4-week-old adult plants, root-regenerated calluses and mixed-stage flowers of A. thaliana(Col-0) were used to extract total RNA using the trizol reagent (Invitrogen). Samples of 20 μg total RNA were resolved in a 15% polyacrylamide/1x TBE/8 M urea gel and blotted to a zeta-probe membrane (BioRad). DNA oligonucleotides with the exact complementary sequence to candidate miRNAs were end-labeled with \[γ-^32^P\]ATP and T4 polynucleotide kinase (New England Biolabs) to generate high specific activity probes. Hybridization was carried out using the ULTRAHyb-Oligo solution according to the manufacturer\'s directions (Ambion), and signals were detected by autoradiography. Finding MPSS evidence for miRNA candidates ------------------------------------------ To obtain genomic regions corresponding to miRNA precursors, we extracted 500 nucleotides upstream and downstream of every genomic locus of all known and predicted ArabidopsismiRNAs. If an intergenic region encoding a miRNA had fewer than 500 nucleotides on either side of the miRNA locus, sequences were extracted up to the neighboring gene. Two ArabidopsisMPSS datasets were used in this study: a MPSS database from abscisic acid (ABA)-treated plants, and plants with elevated levels of endogenous cytokinin \[[@B43],[@B44]\] and a second public MPSS dataset produced by the Meyes laboratory at the University of Delaware, which covers gene-expression information for five Arabidopsistissues at different developmental stages - around 10-week-old active growing calluses initiated from seedlings, mixed-stage buds and immature flowers, 14-day-old leaves, 14-day-old roots and 24- to 48-h post-fertilization siliques \[[@B42]\]. miRNA target gene prediction and 5\' RACE ----------------------------------------- miRNA target gene prediction was performed by aligning miRNA sequences with target mRNA sequences using the TimeLogic implementation of the Smith-Waterman nucleotide-alignment algorithm. Sequences of known and predicted ArabidopsismiRNAs and their O. sativaorthologs were used as query datasets. mRNA sequences of the Arabidopsisand O. sativaannotated genes were used as target datasets. Gaps were allowed in the pairing of miRNA and their target mRNAs. Mismatches were preferred over gaps by assigning higher penalties to gaps in the alignment algorithm. Consecutive gaps were preferred over scattered individual gaps by assigning higher penalties to gap opening than to gap extension. The top 500 putative hits from ArabidopsismiRNA target list and their O. sativaortholog target list were compared. For each mRNA hit of an ArabidopsismiRNA, if a rice ortholog from the same gene family was also found among the top 500 rice miRNA hits, the ArabidopsismRNA hit was selected as a putative miRNA target. Tissue comparison (reference vs flower) microarray data used for target gene validation were downloaded from TAIR \[[@B46]\]. mRNA samples from whole plants and flowers were used as reference and sample probes in the microarray hybridization. To identify the products of miRNA-directed cleavage we used the First Choice RLM-RACE Kit (Ambion) in 5\' RACE experiments, except that we used total RNA (2 μg) for direct ligation to the RNA adaptor without further processing of the RNA sample. Subsequent steps were according to manufacturer\'s directions. Oligonucleotide sequences for PCR amplification of At3g26810, At1g12820 and TIR1 (At3g62980) are available upon request. miRNA clustering and alignment ------------------------------ Predicted miRNAs were compared to known ArabidopsismiRNAs using the MEME motif-searching software and Smith-Waterman gapped local alignment to identify homologs of known miRNAs. Pairs of aligned sequences were grouped by transitive closure, and multiple alignments were generated with ClustalW \[[@B52]-[@B54]\]. The multiple alignment output was manually curated. Free energy calculation ----------------------- We used mfold program to calculate the free energy (ΔG) of predicted miRNA::mRNA pairs. For each miRNA::mRNA pair, the miRNA sequence was linked by \'LLL\' to the target mRNA sequence. The \'LLL\' linker sequence tells the mfold program to treat the miRNA and target sequence as two separate RNA sequences for energy calculation \[[@B38]\]. Note added in proof =================== During revision of this manuscript, three groups \[[@B55]-[@B57]\] reported novel ArabidopsismiRNAs, some of which are included among the predicted miRNAs in this work, confirming the validity of our approach. Additional data files ===================== The following additional data are available with the online version of this paper: the complete list of predicted miRNAs (Additional data file [1](#s1){ref-type="supplementary-material"}); a full list of the secondary structures of predicted precursors of ArabidopsismiRNA candidates and their rice orthologs (Additional data file [2](#s2){ref-type="supplementary-material"}); MPSS evidence for known and predicted ArabidopsismiRNAs (Additional data file [3](#s3){ref-type="supplementary-material"}); a complete list of predicted target mRNAs and their pairing with miRNA sequences (Additional data file [4](#s4){ref-type="supplementary-material"}). Supplementary Material ====================== ::: {.caption} ###### Additional data file 1 The complete list of predicted miRNAs ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 2 A full list of the secondary structures of predicted precursors of ArabidopsismiRNA candidates and their rice orthologs ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 3 MPSS evidence for known and predicted ArabidopsismiRNAs ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 4 a complete list of predicted target mRNAs and their pairing with miRNA sequences ::: ::: {.caption} ###### Click here for additional data file ::: Acknowledgements ================ We thank Michael Zuker for his discussion and suggestion on RNA structures. We appreciate the generous help from Christoph W. Sensen and Paul Gordon at the University of Calgary for the use of their TimeLogic board for executing the Smith-Waterman sequence alignment. We thank Gene Myers for guidance on efficient hairpin search. J.L.R. is a PEW Latin American Fellow. This research was supported in part by NSF grant DBI 998-4882 to T.G. and NIH grants GM 62529 to T.G. and GM 44640 to N-H.C. Corresponding author T.G. can be reached at gaasterland@ucsd.edu as well as gaasterland@rockefeller.edu. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flowchart of the ArabidopsismiRNA prediction procedure. The number of predicted miRNA candidates and potential miRNA precursors (hairpins) is shown in blue bars. The number of known ArabidopsismiRNAs included in each prediction step is shown in parentheses. Known ArabidopsismiRNAs rejected by each prediction step are shown in red boxes. ::: ![](gb-2004-5-9-r65-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Northern blot analysis of predicted miRNAs. Total RNA (20 μg) from 2-day-old seedlings (Se), 4-week-old adult plants (Pl), root-regenerated calluses (Ca), and mixed-stage flowers (Fl) was resolved in a 15% polyacrylamide/8 M urea gel for northern blot analysis. (a)Hybridization signal from confirmed miRNAs. (b)Antisense and sense oligonucleotides (indicated by AS and S, respectively) were used to confirm the polarity of miR417. (c)Hybridization signal for miR158 and 5S rRNA as indicated. The number next to each panel represents the position of RNA markers in nucleotides. In all cases the number in parentheses indicates the time of film exposure in days. ::: ![](gb-2004-5-9-r65-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Putative secondary structures of selected miRNA precursors. (a-c)Secondary structures of predicted precursors of ArabidopsismiR393a, miR416 and miR396b, respectively. (d)pri-mir structure of proposed O. sativahomolog of ArabidopsismiR396b shown in (c). Sequences of mature miRNAs are marked with a red box. ::: ![](gb-2004-5-9-r65-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Comparison of predicted miRNAs with sequences in the ArabidopsisASRP database. Sequences from the ASRP database are named as \'sRNA\' followed by clone numbers. Sequences of predicted miRNAs and sequences from ASRP database are shown in red; miRNA sequences extended according to cloned RNA sequences are in black. The final miRNA sequences reported in Additional data file 1 are marked with asterisks. ::: ![](gb-2004-5-9-r65-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Clusters of predicted miRNAs with known ArabidopsismiRNAs. Identical nucleotides in predicted (underlined names) and known ArabidopsismiRNAs are highlighted in red; differences are highlighted in black; adjacent genomic sequences are shown in black in parentheses. NB indicates miRNAs whose expression was detected as positive by northern blot hybridization; ASRP indicates sequences present in the ASRP database. ::: ![](gb-2004-5-9-r65-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Validation of a miRNA-cleaved mRNA target. (a)Northern blot analysis of miR393a showing its expression pattern. Samples are identical to those in Figure 2b. (b)The 5\' RACE product for the predicted target gene At3g26810 amplified by PCR is shown in the ethidium bromide-stained agarose gel. (c)The 5\' end of the cleaved product determined by sequencing is indicated by an arrow in the miRNA:mRNA base-pairing diagram, along with the number of clones analyzed. The gene structure of At3g26810 and location of the miRNA-binding site are shown at the bottom. ::: ![](gb-2004-5-9-r65-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### miRNAs verified by northern blot hybridizations and their supporting evidence ::: miRNA name Sequence NB MPSS ASRP ------------ ------------------------- ---- ------ ------ miR171b `CGAUUGAGCCGUGCCAAUAUC` \+ \+ NA miR413 `AUAGUUUCUCUUGUUCUGCAC` \+ \+ NA miR414 `UUCAUCUUCAUCAUCAUCGUC` \+ \+ NA miR415 `AACAGAGCAGAAACAGAACAU` \+ \+ NA miR416 `UGAACAGUGUACGUACGAACC` \+ NA NA miR417 `UGAAGGUAGUGAAUUUGUUCG` \+ NA NA miR393a `UCCAAAGGGAUCGCAUUGAUC` \+ NA NA miR418 `UAAUGUGAUGAUGAACUGACC` \+ \+ NA miR419 `UUAUGAAUGCUGAGGAUGUUG` \+ \+ NA miR169b `CAGCCAAGGAUGACUUGCCGG` \+ NA \+ miR396b `UUUCCACAGCUUUCUUGAACU` \+ NA NA miR420 `UAAACUAAUCACGGAAAUGCA` \+ \+ NA miR169g\* `UCCGGCAAGUUGACCUUGGCU` \+ NA NA miR172b\* `GCAGCACCAUUAAGAUUCAC` \+ \+ \+ NB, northern blot hybridization; MPSS, massively parallel signature sequence; ASRP, sequence present in the ArabidopsisSmall RNA Project database; NA, data not available. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Analysis of predicted miRNA target functions using GO annotation ::: Molecular function Number of putative targets ---------------------------------- ---------------------------- Antioxidant activity 17 Nucleic acid binding 80 Catalytic activity 152 Enzyme regulator activity 4 Signal transducer activity 51 Structural molecule activity 1 Transcription regulator activity 129 Translation regulator activity 27 Transporter activity 41 Targets with GO annotation 254 Total predicted targets 371 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Family specificity of putative miRNA-targeted transcription factors ::: Transcription factor gene family Predicted number of proteins\* Predicted number of miRNA targets Percent members targeted^†^ ---------------------------------- -------------------------------- ----------------------------------- ----------------------------- ---- ---- ------- MYB superfamily 190 6 3 10 22 11.6% bHLH 139 46 25 8 3 2.2% HB 89 103 84 9 4 4.5% MADS 82 2 2 4 4 4.9% bZIP 81 21 25 21 4 4.9% CCAAT 36 7 7 6 1 2.8% AP2 14 0 0 0 6 42.9% NAC 109 0 0 0 10 9.1% WRKY 72 0 0 0 1 1.4% GRAS 32 0 0 0 9 28.1% SBP 16 0 0 0 8 50.0% \Data in this column are taken from \[58\]. ^†^The percentage of transcription factors in each family targeted by miRNA in Arabidopsis. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Flower microarray expression data for putative targets of miRNAs identified by northern blot hybridization ::: miRNA Target description Target ID Expression change --------- ----------------------------------------------- ----------- ------------------- miR414 DEAD box RNA helicase At1g20920 -1.10 fold F-box protein family At1g15670 -1.06 fold At2g44130 -1.98 fold SNF2domain/helicase domain-containing protein At3g42670 -1.29 fold Nucleosome assembly protein At4g26110 -1.37 fold miR393a F-box protein family At3g62980 -1.07 fold At3g26810 -1.67 fold miR419 Histidine kinase At2g01830 -3.35 fold miR169b CCAAT box binding factor At5g12840 -2.60 fold miR396b Transcription activator At2g36400 -1.51 fold At4g37740 -2.08 fold ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Summary of evidence for predicted miRNAs ::: Category Number of miRNA candidates ----------------------------------------------------------------------- ---------------------------- Total predicted miRNAs 95 Previously known miRNAs included 12 Newly predicted miRNAs 83 Northern blot positive 12 (14 total) cDNA (ASRP) 8 MPSS 16 Known ArabidopsismiRNA homologs 8 Northern blot and MPSS 8 ASRP and MPSS 3 Unique miRNAs with northern blot hybridization, ASRP or MPSS evidence 25 miRNAs with functionally conserved targets in Arabidopsis*and rice 77 :::	PubMed Central	2024-06-05T03:55:48.501062	2004-8-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522872/", "journal": "Genome Biol. 2004 Aug 31; 5(9):R65", "authors": [ { "first": "Xiu-Jie", "last": "Wang" }, { "first": "José L", "last": "Reyes" }, { "first": "Nam-Hai", "last": "Chua" }, { "first": "Terry", "last": "Gaasterland" } ] }
PMC522873	Background ========== Estrogens are involved in a number of vertebrate developmental and physiological processes. Human and animal studies have revealed the roles of estrogen receptor (ER) in female and male sexual development and behavior, reproductive functions, and the regulation of the neuroendocrine and cardiovascular systems and bone metabolism \[[@B1]\]. Molecular characterizations of breast tumors and epidemiological studies have also shown important roles for estrogens and ERs in the genesis, progression, and treatment of breast cancers \[[@B2],[@B3]\]. Two ER subtypes, ERα and ERβ, are known to mediate estrogen signaling; and they function as ligand-dependent transcription factors \[[@B4]\]. After traversing the cellular membrane, estrogens bind to the receptors, leading to receptor activation. ERs interact with cis-regulatory elements of target genes either directly by binding to previously described conserved estrogen response elements (EREs; 5\'-GGTCANNNTGACC-3\', where N is any nucleotide) or indirectly by associating with AP-1 and Sp1 transcription factor complexes and their respective binding sites \[[@B5]-[@B9]\]. Co-activators and co-repressors form complexes with ERs and are involved in regulating estrogen responses \[[@B10]\]. The cyclical turnover of ER and transcriptional complexes at the regulatory elements of target genes also presents an additional regulatory mechanism \[[@B11]-[@B13]\]. Tissue-specific distribution of co-regulators, associated transcription factor complexes, and receptor subtypes and splice variants are potential mechanisms for the observed pleiotropic effects of estrogens \[[@B14]\]. At the molecular level, the consequence of ER activation appears to be alterations in transcriptional activity and expression profiles of target genes. A number of genes, including those for trefoil factor 1/pS2, cathepsin D, cyclin D1, c-Myc and progesterone receptor, are positively regulated by ERα \[[@B15]-[@B20]\]. Transcriptional repression by ERs has been documented but is not as well studied or understood. Microarray experiments have been carried out, particularly in breast tumor cell lines, to study alterations in gene-expression profiles in response to estrogen treatment \[[@B21]-[@B27]\]. Many key issues remain to be addressed, however, using these initial inventories of responsive genes, including overall conservation of responses across cell lines, in vivorelevance in breast tumors, and cis-regulatory element mapping and molecular characterization and confirmation of the interaction between ER and putative target genes. In this study, we took a combinatorial approach to ERα target gene discovery and characterization by using high-density DNA microarrays to obtain a global gene-expression profile of hormone response in ERα-positive (EPα^+^) breast tumor cells. This included drug treatments that interrogate ER-mediated and translation-independent regulation, integration of additional in vitroestrogen-response data and human breast tumor sample gene-expression data for candidate gene validation and identification of relevant in vivotargets, computational binding site modeling and promoter analysis to map putative ER-binding sites, and chromatin immunoprecipitation (ChIP) to characterize the interaction between ER and the regulatory elements of candidate target genes. Here we present our findings and discuss the insights they provide into the genome-wide architecture of the ER-mediated transcriptional regulatory network and its conservation in cell lines, breast tumors and through evolution. Results ======= Global gene expression profile of estrogen response --------------------------------------------------- High-density DNA microarrays are powerful tools that simultaneously determine the transcriptional profiles of thousands of genes and are especially well suited for studies of transcription factor function. Previous efforts to determine changes in gene expression profiles following hormone treatment in MCF-7 \[[@B21]-[@B25],[@B27]\] and ZR-75-1 \[[@B26]\] ER^+^breast carcinoma cell lines have yielded a number of novel estrogen-responsive genes and demonstrated the utility of such genome-scale technologies in studying estrogen biology. These earlier studies, however, only included anti-estrogen and cycloheximide (CHX) treatments either at a limited number of time points or only in validation assays for a handful of putative responsive genes. Therefore, to map more comprehensively the transcriptional regulatory network regulated by ER and to generate data in an additional ER^+^breast tumor cell context for cross-cell line analysis, we treated the estrogen-dependent T-47D ER^+^breast cancer cell line with 17β-estradiol (E2) and with E2 in combination with either the pure anti-estrogen ICI 182,780 (ICI) or the protein synthesis inhibitor CHX and performed high-resolution time-course gene-expression analyses (see Figure [1a](#F1){ref-type="fig"} for treatments and time points) using spotted oligonucleotide (60-mers) microarrays containing probes representing around 19,000 human genes. The concentrations of E2 (1 nM) and ICI (10 nm) used in this study were sufficient to respectively drive and inhibit hormone-responsive cell proliferation. T-47D cells differ in karyotypic abnormalities and nuclear receptor co-regulator expression levels from cell lines previously used - MCF-7 and ZR-75-1 - but have the advantage of expressing ER at more physiologic levels \[[@B28],[@B29]\]. Samples were harvested on an hourly basis for the first 8 hours (0-8 hours) following hormone treatment and bi-hourly for the next 16 hours (10-24 hours) for a total of 16 time points surveyed (Figure [1a](#F1){ref-type="fig"}). Estrogen-responsive genes were determined by statistical analysis of expression ratios in E2-treated samples versus the mock-treated controls using a two-tailed paired t-test with p-value cutoff. In addition, the genes were filtered for at least a 1.2-fold change in the same direction in three or more time points. Our choice of data-selection criteria was informed by the observed expression levels and profiles of known hormone-responsive and ER target genes such as the progesterone receptor, cathepsin D and stanniocalcin 2. Responsive genes were selected for E2 responsiveness (p\< 0.052) and filtered for ICI sensitivity (p\< 0.057) and CHX insensitivity (p\> 0.24) by comparing E2-treated samples with E2+ICI and E2+CHX samples to isolate putative ER downstream targets and direct targets, respectively. Figure [1b](#F1){ref-type="fig"} summarizes the statistics of the selection process. Expression profiles of estrogen-responsive genes were visualized by Eisen clustergrams and the genes were sorted by hierarchical clustering \[[@B30]\]. For estrogen-responsive genes (Figure [2a](#F2){ref-type="fig"}), the expression profiles clustered into two groups: genes that are upregulated by hormone treatment (Figure [2a](#F2){ref-type="fig"}, red) and those that are downregulated (Figure [2a](#F2){ref-type="fig"}, green). Of the responsive genes, 58.5% (226/386) were upregulated following estrogen treatment. Figure [2b](#F2){ref-type="fig"} shows the expression profiles of the 137 genes specifically regulated by ER (defined by being responsive to estrogen and blocked by ICI treatment, Figure [2b](#F2){ref-type="fig"}, right panel). These genes cluster in a similar fashion as the estrogen-responsive genes. A notable finding is that ER-regulated genes, as determined by E2 response and ICI sensitivity, account for only 35.5% (137/386) of all estrogen-responsive genes, suggesting the possibility of ER-independent signal transduction mechanisms in mediating transcriptional responses to hormone exposure. Interestingly, ICI appears to have a greater effect on E2-upregulated than downregulated genes as these upregulated genes account for 71.5% (98/137) of the ICI-sensitive subset. Eighty-nine primary response genes constituted the putative ER direct targets - defined as responsive to hormone treatment, sensitive to ICI but not affected by CHX treatment (Figure [2c](#F2){ref-type="fig"}, right panel). From these observations the number of direct target genes involved in initiating hormone response in breast tumor cells might represent only 0.47% (89/18,912) of the genes in the human genome. The list of putative ER target genes is presented in Table [1](#T1){ref-type="table"}, and the complete listing of E2-responsive genes is given in Additional data file 4. Genes previously shown to be ER targets are shown in bold type (see Table [1](#T1){ref-type="table"}). We note that five of the direct target genes on our list correspond to the results presented in the only other published microarray study to include anti-estrogen and CHX treatments as part of the experimental design \[[@B26]\]. The discrepancy between the two datasets is likely to be due to the differences in cell lines, experimental designs, array platform and the filtering parameters used for selecting target genes. For example, the Soulez and Parker study \[[@B26]\] utilized ZR-75-1 cells tested at only two time points (6 and 24 hours). Moreover, the Affymetrix HuGeneFL arrays used in that study contained probes for 5,600 genes as compared to the nearly 19,000 genes on our arrays, and represented early array technology with known limitations. Similarly, the absence of known target genes TFF1/pS2 and cyclin D1 in our list of direct targets is probably due to differences in the transcriptional cofactors and genomic abnormalities in the T-47D cells. The absence of the c-MYCproto-oncogene on our list is probably because the probe was not represented in the arrays used in our study. To control for the confounding direct effects of ICI and CHX, independent of E2, we also treated the cells with only ICI for 2, 8, 12 or 24 hours, or only CHX for 2 or 8 hours. CHX treatment partially obscured the responses in 4 of 89 genes (4.5%) that met the selection criteria for putative direct target genes by inducing a detectable E2-like effect following either CHX, E2, or E2+CHX treatments (see Additional data files 5 and 6). As this result does not rule out these genes as direct targets, we included them in the direct target list, but noted this caveat in Table [1](#T1){ref-type="table"}. CHX treatment alone had the opposite effect to E2 treatment in 32 (8.3%; 32/386) of the E2-responsive genes. However, in the presence of the hormone and the drug together (E2+CHX), CHX did not antagonize the E2 response of these genes and therefore did not affect the selection of putative direct targets. ICI treatment alone elicited an unexpected E2-like response (that is, identical response following either E2, ICI or E2+ICI treatments) in nine (2.3%; 9/386) E2 responsive genes. Because these nine genes did not meet the selection criteria for ICI antagonism, the putative direct target genes were also not affected. Thus, the independent effects of ICI and CHX did not substantially alter the final gene list or conclusions in our analysis. Comparison of T-47D and MCF-7 estrogen-response profiles -------------------------------------------------------- A number of breast cancer cell lines have been used in in vitrostudies of estrogen responses, but most of the data, including those from published microarray studies, have come from experiments using MCF-7 cells. Therefore, we were interested in comparing the expression profiles in response to hormone treatment between MCF-7 cells and the T-47D cells used in our studies. For our analysis, we obtained a publicly available MCF-7 hormone and tamoxifen response dataset \[[@B31]\] from the Stanford Microarray Database \[[@B32]\]. Using Unigene cluster IDs from build 166 as the common identifiers between the two datasets, we extracted expression data from 104 of 137 T-47D ER-regulated genes (Figure [3a](#F3){ref-type="fig"}) that were also present in the MCF-7 dataset. For genes with multiple entries in the MCF-7 data, the entry with either the most complete data or with similar expression profiles to the T-47D results was selected for analysis. Overall, the results from the MCF-7 experiments correspond to the majority (64%; 66/103) of expression profiles of responsive genes obtained in the T-47D cells as defined by same direction changes in the available data points in MCF-7 cells following hormone treatment (see Figure [3a](#F3){ref-type="fig"}). Using stringent selection criteria for the MCF-7 data for E2 response and sensitivity to tamoxifen treatment (see Materials and methods), we found 24 genes that can be defined as ER regulated in both MCF-7 and T-47D datasets. Remarkably, there was high concordance between the two studies with 23 out of 24 (96%) of the genes showing concordance in their expression response to estrogen (Figure [3b](#F3){ref-type="fig"}). In contrast, there was little concordance in microarray datasets from unrelated stem-cell studies despite use of similar experimental systems and identical array platforms \[[@B33]\]. These findings were further validated by the many overlaps between the genes identified in this investigation and the estrogen-responsive genes reported by Frasor and colleagues (data not shown) \[[@B23]\]. The similarities we observe between the two ER studies with different experimental designs and array platforms suggest that the two ER^+^cell lines share common estrogen-response pathways. Differential expression of putative ER-regulated genes in breast tumors ----------------------------------------------------------------------- A key question we wished to address was whether the in vitroobservations in cell lines reflected biological significance in vivo. To address this, we explored the association between the ER-regulated genes identified in our in vitroanalysis and the ER status-associated expression profiles in breast tumor samples. We hypothesized that putative ER target genes should be differentially expressed in breast tumors in an ER status-dependent manner. For example, pS2/TFF1 and cyclin D1, both upregulated by estrogen treatment in MCF-7 cells, were shown to be expressed at higher levels in ER^+^tumors \[[@B34],[@B35]\]. A number of breast cancer microarray studies have shown that ER status remains the most important prognostic marker and tumor classifier. Expression data from six breast cancer microarray studies (L.D.M., B.M.F. Mow, L.A.V., and E.T.L., unpublished work and \[[@B36]-[@B40]\]) were mined for genes that were differentially expressed (p-value \< 0.01 false discovery rate, ER^+^vs ER^-^) in human breast tumor samples with respect to ER status. Of the 137 ER-regulated genes (E2 responsive, ICI sensitive) identified in the T-47D study, 44 genes were differentially expressed in at least one breast cancer study (Figure [4](#F4){ref-type="fig"}). The 44 ER-regulated genes represent only about 1% (44/3811) of the 3,812 ER-status-associated genes that met the selection criteria (p\< 0.01 in one or more studies), suggesting that the estrogen-responsive pathways represent only a minor part of the ER-status-associated transcriptome in breast tumors. This is similar to observations made previously by Meltzer and co-workers \[[@B22],[@B39]\]. However, there appears to be a significant enrichment of ER-regulated genes within the ER-status-associated genes compared to the frequency of these ER-regulated genes represented in the microarray used in our study (1.15%, 44/3,811 vs 0.72%, 137/18,912, p= 0.006 by chi-square analysis). To compare the expression profiles of responsive and differentially expressed genes, we plotted the average relative expression ratios of each gene (ER^+^/ER^-^) across all samples from the breast cancer studies (Figure [4](#F4){ref-type="fig"}). There was surprising concordance (70.5%; 31/44) between the estrogen-responsive genes identified in T-47D cells and genes differentially expressed in breast tumors. For example, genes upregulated by hormone treatment (Figure [4](#F4){ref-type="fig"}, left panel, red) were also overexpressed in ER^+^breast tumors (Figure [4](#F4){ref-type="fig"}, right panel, red). We noted a subset (29.5%; 13/44) of genes that exhibited opposite responses following estrogen treatment in vitroas compared to the ER-status-associated expression in tumors. These 13 genes that are discordant between cell line and tumor data were, however, consistent across the two cell lines (T47-D and MCF-7). This suggests context-dependent regulation of some downstream pathways, which is likely to be different between primary tumors and experimental cell lines. Taken together, we note that these in vitrovalidated estrogen-responsive genes are also differentially expressed in ER^+^primary tumors, and may therefore have direct biological and clinical significance. Computational modeling and predictions of ER-binding sites ---------------------------------------------------------- Previous studies have identified the consensus ERE and the AP-1- or Sp1-binding sites in DNA as possible target motifs for. This would suggest that the 89 direct responding genes should be enriched for these binding motifs within the transcriptional control regions. To further explore this, we computationally extracted sequences flanking (-3,000 to +500) the transcriptional start site (TSS, see Materials and methods section), defined as the most 5\' nucleotide of the reference transcript in the NCBI RefSeq database, of candidate genes and queried them for potential ER-binding sites. The size and locations of the sequences flanking the start sites were selected because most of the characterized ER-binding sites have been mapped to these regions in known target genes \[[@B5]\]. For binding-site predictions we used our previously described ERE model \[[@B41]\] and AP-1 and Sp1 binding-site position weight matrices from the TRANSFAC database \[[@B42]\]. We also included the binding site for the GATA1 transcription factor as a negative control as it is not known to be involved in ER binding. Model sensitivities for all the sites surveyed were set at the established optimal setting for the ERE model of 83% sensitivity in detecting known binding sites in the training data for the models. Figure [5](#F5){ref-type="fig"} shows the performance of the ERE (Figure [5a](#F5){ref-type="fig"}), AP-1 (Figure [5b](#F5){ref-type="fig"}), Sp1 (Figure [5c](#F5){ref-type="fig"}), and GATA1 (Figure [5d](#F5){ref-type="fig"}) binding-site models. The y-axis for each graph represents the relative frequency of binding-site prediction as determined by the fraction of genes with predicted binding sites over the total number of genes queried; the x-axis represents the number of most significant genes investigated, ordered by statistical significance, for each of the groups of genes (see Materials and methods). Since short binding site motifs are ubiquitous in the human genome, we asked whether there was enrichment of such response elements in the 3.5 kilobase (kb) upstream windows of responsive genes as compared to unresponsive genes. Enrichment for each motif is represented by a divergence of the relative frequencies of binding-site predictions for putative target genes (Figure [3](#F3){ref-type="fig"}, solid lines) and non-responsive genes (Figure [3](#F3){ref-type="fig"}, fragmented lines). For ERE predictions, we observed a threefold enrichment of putative sites in the 10 most significant primary response genes as compared to the most non-responsive controls (Figure [3a](#F3){ref-type="fig"}), and twofold and approximately 70% enrichment for the 25 and 50 most significant genes, respectively. Overall, the enrichment of ERE sites in putative ER direct target genes is statistically significant (p= 0.0027). The enrichment of putative Sp1 sites in the target genes was more modest but did not reach statistical significance (12.5% enrichment for the 10 most significant target genes; p= 0.085). This is expected as Sp1 sites are quite common in the human genome and additionally function in general transcriptional regulation. We did not observe any enrichment of AP-1 sites (p= 0.66) or the negative control GATA1 sites (p= 0.51). These findings suggest that the ERE is the major response element mediating the specific regulation of ER target genes on a whole-genome scale. We also surmised that although Sp1 and AP-1 binding sites are known to facilitate ER functions in some target genes they are not used as a common ER-targeted cis-regulatory element within the human genome, at least not sufficiently to distinguish target genes from non-responsive genes. To determine the conservation and potential functionality of the predicted EREs, we also examined the same 3.5 kb window in the 5\' upstream regions of mouse orthologs of the 89 putative human ER target genes. Seventy-two human-mouse orthologous gene pairs were extracted from the Mouse Genome Database \[[@B43]\] and the regulatory regions demarcated and analyzed for potential EREs as described for the human sequences (see Materials and methods). We then compared the ERE predictions from the two organisms for the following features: conservation of the core ERE half-sites (GGTCANNNTGACC), excluding the flanking purine bases, between the two most similar sequences when multiple EREs are predicted in either organism; conservation of the 20 bases flanking the 5\' and 3\' ends (40 bases total) of the predicted EREs; and the distance between the binding-site sequences and the TSS. The statistics of our analysis is summarized in Figure [6a](#F6){ref-type="fig"}. Of the orthologous mouse-human pairs, 81% (58/72) have at least one ERE prediction and 22 (31%; 22/72) gene pairs have ERE predictions in both organisms. However, of the human direct target genes, 29% (21/72) have no EREs upstream of the mouse orthologs. Conversely, 21% (15/72) of the mouse genes with EREs have no ERE upstream of their human orthologs. Of the 22 gene pairs that have ERE predictions in both organisms (see Venn diagram in Figure [6b](#F6){ref-type="fig"}), only four have perfect conservation of the core ERE sequences (Table [2](#T2){ref-type="table"}). These four perfectly conserved ERE pairs also have the highest conservation in their flanking sequences (average identity = 74%) and the smallest difference in the relative positions of binding sites (average difference Δd= 469 bases) between the human and the mouse sequences. In fact, the relative positions of the conserved EREs only differ by an average of 52 bases if the predicted EREs for GREB1(human, NM\_014668; mouse, NM\_015764), which differed by 1.7 kb in their relative position, were excluded from the analysis. For the ERE mouse-human pairs with one or more base deviations in their core sequences, there is little conservation in the flanking sequences and in the relative positions of predicted EREs (see Table [2](#T2){ref-type="table"}). These findings indicate that although the ERE motif is conserved through evolution, specific EREs found in the 5\' regulatory regions of target genes are rarely conserved. They also suggest potential differences in the molecular mechanisms of ER function and in the repertoire of target genes between human and rodents. In light of this, our inference of ER function in humans from the results obtained from animal studies may warrant a re-evaluation and additional validation. Validation of direct ER target genes by chromatin immunoprecipitation --------------------------------------------------------------------- The genomics and informatics approaches have enabled us to identify genes that meet the conventional definition for ER target genes (for example, responsive to E2, sensitive to ICI, and insensitive to CHX), are conserved in ER^+^breast cancer cell lines and tumor samples, and encode putative ER-binding sites in the promoter regions. Two genes emerged at the top of the list of direct target genes following these analyses. One was for nuclear receptor-interacting protein 1 (NRIP1), also known as receptor-interacting protein 140 (RIP140), first identified as an ER-binding protein and a co-regulator of receptor activity \[[@B44],[@B45]\]. It was subsequently shown to bind and modulate transcriptional activities of other nuclear receptors \[[@B46],[@B47]\]. Previous microarray experiments in MCF-7 and ZR75-1 cells showed that NRIP1transcript levels were raised following estrogen treatment, and its expression dynamics in the presence of anti-estrogens and CHX were consistent with other primary response genes \[[@B23],[@B24],[@B26]\]. In this study, we have also identified NRIP1as a putative ER target gene that is upregulated by E2, sensitive to ICI treatment and insensitive to CHX in T-47D cells. Furthermore, we detected a conserved perfect ERE at around 700 bases upstream of the TSS, indicating a potential ER-binding site and direct regulation by the activated receptor. The other direct target gene - gene regulated by estrogen in breast cancer 1 (GREB1) - was identified in a subtractive hybridization screen for estrogen-responsive genes in MCF-7 cells. It has no known function and does not appear to share significant homology with any other gene in the sequence databases \[[@B48]\]. A perfect ERE was found at around 1.6 kb upstream of the TSS of GREB1and the predicted ERE is also conserved in mouse. Given that both NRIP1and GREB1have been conserved during vertebrate evolution, we compared the 5\' upstream regions from human, chimpanzee, mouse and rat genome sequences to see whether the predicted regulatory element has been conserved in additional murine and primate species. For all of the regions surveyed, we found that the core ERE has been perfectly conserved (Figure [7a](#F7){ref-type="fig"}). In addition, sequences flanking the predicted ERE were also highly conserved, suggesting functionality for these regions. To determine the role of the predicted ERE as an ER-binding site, we performed chromatin immunoprecipitations (ChIPs) using anti-ER antibodies. In addition to the two conserved EREs, we also included two non-conserved EREs from TFF1/pS2(positive control) and ATP-binding cassette, subfamily A, member 3 (ABCA3), a gene related to other ABC transporters that are thought to be involved in cellular lipid transport and which is a putative ER direct target gene as determined in this and a previous study \[[@B26]\]. Forward and reverse primers (Figure [7b](#F7){ref-type="fig"}) flanking the ERE were designed to specifically detect and quantify genomic DNA fragments that co-precipitate with ER in real-time PCR experiments. Following hormone treatments, we did not observe significant enrichment of the negative control actin exon 3 region in anti-ER precipitates as compared to the anti-GST antibody control or the input genomic DNA from the nuclear lysates for all primer pairs tested (Figure [7c](#F7){ref-type="fig"}). In contrast, semi-quantitative PCR analysis (see Materials and methods) of the ChIP products using primers flanking the predicted EREs revealed ER binding to these sites in the absence of estrogen and after hormone treatment (see Figure [7c](#F7){ref-type="fig"}). Furthermore, the binding appeared to be enhanced following estrogen treatment, suggesting a role for activated receptors in mediating the observed transcriptional regulation of these genes. The functionality of the conserved EREs in NRIP1and GREB1was also recently reported in a study of near-consensus EREs in the human and mouse genomes \[[@B49]\]. Discussion ========== We have conducted a genome-wide analysis of E2-responsive genes. Through a strategy of iterative validation using genomics, informatics and experimental biology we have identified and characterized a core set of 89 ER direct target genes out of the 18,912 genes represented on our microarray (0.5%). This set of direct target genes derived from experiments in T-47D cells show very similar behavior in another cell line, MCF-7, and also overlap with genes that can distinguish ER status in human breast cancers. Taken together, these results suggest common underlying mechanisms for ER transcriptional control and define specific genetic components of the ER transcriptional network that are consistent across model experimental and clinical conditions. These results emboldened us to decipher the rules and informational framework underlying ER transcriptional control. The anti-estrogen treatment with the ICI drug and preincubation of the cells with CHX allowed us to identify genes likely to be ER direct targets. We extracted extended promoter regions (- 3,000 bp to +500 bp) and determined potential ER-binding sites by using ERE and AP-1 and Sp1 binding-site models \[[@B41],[@B42]\]. Because transcription factor binding elements occur very frequently in the genome, finding an ERE, AP-1 or Sp1 site only in ER-responsive genes is highly unlikely. Instead, we asked whether the probability of finding an ER-associated response element was significantly higher than that seen in ER-unresponsive genes. Our results, depicted in Figure [5](#F5){ref-type="fig"}, show distinctly that EREs are enriched in the putative direct estrogen-responsive genes (p= 0.0027) but that binding sites for AP1, Sp1 or GATA1 (which we used as a negative control response element) are not. The ERE thus appears to be the predominant ER transcriptional control element. Moreover, despite definitive experiments showing the ability of AP-1 and Sp1 sites to mediate ER responses \[[@B7]-[@B9],[@B47],[@B50]-[@B54]\], our results suggest that their usage is not a common mechanism for specific ER transcriptional control on a genome-wide scale. Previous investigations have uncovered a functional ERE embedded within an Alurepetitive sequence that is frequent in the genome \[[@B55]\]. Inclusion of this AluERE into our analysis, however, dramatically degrades the enrichment of EREs found in direct ER-responsive genes (p= 0.06). This suggests that though such EREs are experimentally functional, they have little impact on the specific ER transcriptional cassette, functioning as no more than \'noise\' in the system. This has been confirmed by negative ChIP data for several Alu-ERE sites (data not shown). These observations highlight the potential confounding factors in genome-wide analysis of functionally relevant response elements. Our use of a 3.5 kb window around the TSS to search for relevant EREs captures the majority of known EREs \[[@B5]\] and represents a liberal survey of 5\' regulatory regions. Despite this, we found that only about 50% of the target genes encode ERE-like sequences (including ERE half-sites) in their promoters. It is possible that ER-binding sites outside this window may be involved in regulating the specific activities of ERs. In support of this, Bourdeu and colleagues very recently described the identification and validation of EREs within DNA 10 kb upstream (relative to TSS) and 5 kb downstream in 5\' regions of a number of human genes \[[@B49]\], indicating the presence of functional enhancer elements outside the region surveyed in our study. In addition, errors in annotating the TSS or additional 5\' exons may account for up to an 8% error rate for TSS determination in known genes and 80% error rate in predicted genes (Y.J. Ruan, E.T.L. and C.L. Wei, unpublished work). Future studies will need to incorporate these information in the ERE analyses. Given that in silicoidentification of EREs does not assure their function in an ER response, we selected three new putative direct ER target genes identified by our stringent criteria for further validation. NRIP1, GREB1, and ABCA3are all genes found to be ER responsive in at least two cell lines; they have a discernable ERE around the TSS, blocked by ICI and not inhibited by CHX, and their expression can discern ER status in breast cancers. Using ChIP we confirmed that the EREs in all three are directly targeted by ER following estrogen stimulation (Figure [7](#F7){ref-type="fig"}). Therefore, our process of ranking by consensus (that is, ranking by likelihood of being a direct target of ER by the number of criteria fulfilled) appears to be a reasonable approach to identify actual direct targets of ER. These target genes suggest potential roles for ER in regulating intracellular signaling pathways that may have an impact on processes in breast and tumor biology. NRIP1was first identified as an ER-binding co-regulator protein and was subsequently found to interact with other nuclear receptors through the nuclear receptor binding motif LXXLL. Kerley and colleagues \[[@B56]\] showed that NRIP1transcript and protein levels were also upregulated by all-transretinoic acid treatment and suggested that NRIP1may facilitate cross-talk between members of the nuclear receptor family. Thus, upregulation of NRIP1by activated ER may not only modulate the estrogen response but also affect the transcriptional activities of other nuclear receptors and the cellular responses to their corresponding ligands. That NRIP1transcript levels were elevated in ER^+^compared to ER^-^breast tumors suggests that the downstream function of other nuclear hormone receptor may be coordinately modulated by elements of the ER transcriptional cascade (see Figure [4](#F4){ref-type="fig"}). ABCA3encodes a member of the ABC transporters that utilize ATP hydrolysis to drive the transport of substrates across the cell membrane; although its substrate is not known, ABCA3 appears to be related to other ABC transporters involved in lipid transport. Levels of ABCA3 protein are highest in lung tissue, and ABCA3 appears to localize to lamellar bodies of alveolar epithelial cells that are highly enriched in phosphatidylcholine \[[@B57]\]. These observations provide potential links between ER activation and alterations in phospholipid levels during breast epithelial cell differentiation and transformation. GREB1, however, is a gene of unknown function and is unrelated to any other known gene. Its overexpression in ER^+^breast tumors and its evolutionary conservation suggest a central role for this gene in ER signaling and breast tumor biology. Of note, 21% (19/89) of the putative target genes identified in this study have no known biological functions (see Table [1](#T1){ref-type="table"}). One strategy used in assessing cis-regulatory elements in the genome has been to map conserved segments in non-coding regions upstream of TSSs. Using the three genes above (NRIP1, ABCA3and GREB1) as rigorously tested direct targets of ER regulation and a well-studied ER direct target, TFF1/pS2, we assessed the evolutionary conservation of the validated upstream EREs between human and mouse homologs. Interestingly, we found highly conserved EREs (including flanking regions) only for NRIP1and GREB1. ABCA3and TFF1/pS2both have upstream functional EREs in the human genes but not in their mouse orthologs (Figure [7b](#F7){ref-type="fig"}). We then extended this search for evolutionary conservation to the remaining 89 putative human ER direct target genes. Surprisingly, we found that in the majority of mouse-human orthologous pairs, the ERE core sequences, flanking regions and position relative to the TSS are not conserved: only 4 out of the testable 72 (6%) orthologous pairs examined showed conservation of ERE sequences between the human and mouse genes. This is remarkable given the 84.7% \[[@B58]\] identity between mouse and human sequences within coding regions. Taken together, our results suggest that the evolution of transcriptional control through cis-regulatory mechanisms must have different mutational rates or mechanisms, and may have undergone different selection pressures from those imposed on coding sequences. Moreover, the low level of conservation in the EREs of estrogen-responsive genes between mouse and human suggest two consequences: first, that the core physiologic estrogen effects such as sex differentiation/mammary gland development may be mediated by a small set of highly conserved and similarly regulated ER-responsive genes; and second, that there might be significant differences between downstream estrogen effects between mouse and human. We suggest the relevance of many of these estrogen-response genes to breast tumor biology by showing significant similarities between estrogen-induced expression profiles in MCF-7 cells and the behavior of these genes in ER^+^tumors from a database of six breast cancer microarray studies \[[@B36]-[@B40]\]. Not unexpectedly we observed that the number of direct estrogen-responsive genes was small in comparison to the overall number of genes that define the ER^+^breast tumors, suggesting that the estrogen-responsive pathways account for only a portion of the receptor-positive molecular signature, an observation also noted by others \[[@B22]\]. Nevertheless, taken together, it appears that the ER^+^status of primary breast cancers can be accounted for by concordant effects of an activated ER. Interestingly, a number of these differentially expressed genes (around 30%) were expressed in the opposite direction in the cell lines compared to the tumor consensus. We speculate that this may be due either to consistently different profiles of ER cofactors or to an intense expression signature in tumor-associated stromal cells that is opposite to that of the cancer cells. Nevertheless, those genes that are estrogen-responsive in cell lines and differentially expressed in ER^+^tumors represent the most promising candidates for further functional analysis. In summary, we have presented an integrated strategy for discovering and characterizing ER target genes, response elements and the transcriptional regulatory network downstream of ER activation. With this approach, we uncovered a universal set of genes that describe the most direct effects of ER and operate across multiple in vitroand in vivosystems. On examination, this core direct target gene list does not predict a unified biological process controlled by ER. Instead, the gene functions would predict a pleiotropic cellular response. By further in silicoanalysis of the promoter regions, we observed minimal conservation in the cis-regulatory region of the direct estrogen-response genes between humans and mice. This raises the intriguing possibility that the evolutionary processes governing the configuration of transcriptional regulation will be different from those affecting the functional domains of genes. Moreover, we predict that the estrogen response in the mouse will differ significantly from that in the human, but that a small set of ER direct target genes that are highly conserved in their cis-regulatory regions will act as the key effectors of evolutionarily important core ER functions such as sex differentiation. Conclusions =========== Estrogen responses in human breast tumor cells appear to be mediated by a relatively small conserved core set of ER-regulated genes. Examination of the cis-regulatory regions of putative target genes within this core set revealed the enrichment of the ERE sequence motif but not other known ER-binding sites. Of all the predicted EREs in human direct target genes, only a handful (6%) appear to be conserved in mouse orthologs, although both conserved and non-conserved predicted EREs were shown to bind ER in human cell lines. Taken together, these findings suggest the potential for species-specific mechanisms and effects in response to hormone exposure. Materials and methods ===================== Cell culture, treatments and RNA extraction ------------------------------------------- T-47D and MCF-7 cells were maintained in DMEM/F12 (1:1) medium (Invitrogen) supplemented with 10% fetal calf serum (FCS) (Hyclone) at 37°C and buffered with 5% CO~2~. For estrogen treatments, cells were washed with PBS and pre-cultured in phenol-red-free DMEM/F12 medium supplemented with 0.5% charcoal-filtered FCS (Hyclone) for 24 h. For time-course experiments, T-47D cells were treated with 1 nM 17β-estradiol (E2; Sigma-Aldrich) or 1 nM E2 + 10 nM ICI 182, 780 (Tocris Cookson) for the amount of time specified. To determine the primary response, cells were treated with 5 μg/ml cycloheximide (CHX; Sigma-Aldrich) for 30 min before the start of estrogen treatment. Control treatments with ICI (2, 8, 12 and 24 h) and CHX (2 and 8 h) alone were also carried out for the times specified and at the same concentrations as above. To extract RNA, cells were washed with PBS, lysed in Trizol (Invitrogen) and samples were harvested by additional phenol-chloroform extraction steps as prescribed by the manufacturer. Microarray analysis of gene-expression profiles ----------------------------------------------- Microarrays were generated by spotting the Compugen 19 K human oligo library, made by Sigma-Genosys, on poly-L-lysine-coated glass slides. Twenty-five micrograms of each sample total RNA and human universal reference RNA (Stratagene) were labeled with Cy5-conjugated dUTP and Cy3-conjugated dUTP (PerkinElmer), respectively, and hybridized to the arrays using protocols established by the Patrick O. Brown Laboratory \[[@B59]\]. Array images and data were obtained and processed using the GenePix4000B scanner and GenePix Pro software (Axon Instruments). Differentially expressed genes were determined using pairwise t-test between matching treated samples and mock-treated controls at each time point and fold-difference cutoff at multiple time points as described and clustered and visualized using the Eisen Cluster and TreeView programs \[[@B30]\]. Gene ontology of putative target genes was derived from annotations made by Compugen. Meta-analysis of breast cancer and cell line microarray data ------------------------------------------------------------ A database containing published and unpublished breast cancer expression data was queried for genes whose expression profiles differentiated ER^+^and ER^-^tumors. Each individual dataset was analyzed independently for differentially expressed genes by calculating the false discovery rate for each gene \[[@B60]\] and setting the p-value filter at less than or equal to 0.01. ER-status-associated genes were then cross-referenced with the in vitroestrogen-responsive genes via the UniGene cluster ID (build 166). The log-transformed average mean-centered expression values for each statistically significant study were used for visualization. Raw in vitroMCF-7 estrogen response data \[[@B31]\] were downloaded from the Stanford Microarray Database \[[@B32]\]. The data were compared directly with the T-47D results or selected for ER regulation by the following selection criteria: first, at least a 1.15-fold change in the same direction in two out of three time points and no conflicting (opposite direction) data in any of the time points; and second, changes in the opposite direction when co-treated with tamoxifen (Tam) for 48 h in one out of the two treatment conditions and no conflicting data in the two treatments. The 1.15-fold cutoff, which differs from the 1.2-fold change for the T-47D data, was selected to capture known E2-responsive genes in this dataset. Promoter sequence extraction and detection of ER-binding sites -------------------------------------------------------------- The LocusLink and RefSeq \[[@B61]\] databases at the National Center for Biotechnology Information (NCBI) were used to identify human and mouse genes and pinpoint their loci within the genome. These annotations were chosen for their comprehensiveness, in terms of number of annotated genes, and their consistency with the current state of NCBI contig databases. Using the TSS, defined as the most 5\' nucleotide in the reference transcript, and the 3\' terminus of the transcript as reference points, we extracted 3 kb upstream and 500 bases downstream of the start sites for binding-site analyses. NCBI human genome sequence build 33 and mouse genome sequence build 30 were used for transcript alignment and genomic sequence extraction. TSS locations annotated in LocusLink and RefSeq may only approximate true start sites because of incomplete information at the 5\' ends of some reference sequences, but we believe that the relatively large (3.5 kb) regions used for our analysis allow for fluctuations in TSS position. Human-mouse ortholog determinations were based on annotations made in the Mouse Genome Database \[[@B43]\]. The four binding-site position weight matrix (PWM) models used were either derived in an earlier study \[[@B41]\] or downloaded from the TRANSFAC (version 6.0) database of transcription factor binding sites \[[@B42]\]. Detection parameters were set on the basis of optimized settings for the Dragon ERE Finder \[[@B41]\] at 83% sensitivity in detecting training data and corresponding settings were made for the other PWMs to have similar sensitivities. Statistical significance of binding-site enrichment between putative target genes and non-responsive genes was determined by Monte Carlo simulations between predictions in defined gene sets and randomly generated genes sets. A set of Monte Carlo simulations was performed to assess the significance of the apparent enrichment of putative EREs between the set of estrogen direct target genes and the non-responsive genes. In each simulation, we randomly generated two sets of genes (equivalent in sizes to the set of direct target and non-responsive genes), plotted the curves accordingly, and calculated the difference between the areas under the two curves. The simulations were performed 100,000,000 times and the fraction of times in the simulations that the random area-difference was at least as large as the observed area difference was reported as the empirical p-value. Most significant direct target genes used in the analysis were ranked by the lowest p-values from analysis of E2-treated and control samples, E2 and E2+ICI samples, and E2 and E2+CHX samples. Non-responsive genes were ranked by highest p-values from the same analysis. Chromatin immunoprecipitation assays ------------------------------------ MCF7 cells were estrogen deprived for 24 h and treated with 100 nM E2 for 45 min before 1% formaldehyde treatment to cross-link the transcription machinery and the chromatin. Immunoprecipitations were carried out overnight with ERα (HC-20) or GST antibodies (Santa-Cruz Biotechnology) and protein A-sepharose beads (Zymed). Washing and extraction protocols were modified from methods described previously \[[@B62]\] and PCR reactions were carried out in a LightCycler (Roche Diagnostics) real-time system. Forty cycles of PCR were carried out on precipitated DNA and control input DNA using the following primer sets: TFF1/pS2ERE: forward CCATGTTGGCCAGGCTAGTC; reverse ACAACAGTGGCTCACGGGGT. NRIP1ERE: forward, TGCTCCTGGGTCCTACGTCT; reverse TCCCCTTCACCCCACAACAC. GREB1ERE: forward AGCAGTGAAAAAAAGTGTGGCAACTGGG; reverse CGACCCACAGAAATGAAAAGGCAGCAAACT. ABCA3ERE: forward, CACCTTCCATCTGTCCAAAG; reverse, CAACCCTGAGGTTTGGGAAC. Actin exon 3 control: forward, AGACCTTCAACACCCCAGCC; reverse, GTCACGCACGATTTCCCGCT. Amplification products were also assayed for specificity by melting-curves analysis at the end of each run. Relative quantifications were carried out by building standard curves for each primer set and using genomic DNA, similar to the input, as the template. Enrichment of ER binding was determined by comparing the relative quantities of anti-ER and control anti-GST products. Additional data files ===================== The following additional data files are available with the online version of this article: the processed raw data for the time course microarray study (Additional data file [1](#s1){ref-type="supplementary-material"}) and replicate 1 (Additional data file [2](#s2){ref-type="supplementary-material"}) and replicate 2 (Additional data file [3](#s3){ref-type="supplementary-material"}) of the control microarray study with the ICI and CHX treatments alone, the complete list of all 387 estrogen-responsive genes described in the article with UniGene cluster numbers from build 166 (Additional data file [4](#s4){ref-type="supplementary-material"}), expression profiles of ICI and CHX responsive genes identified in the control experiments (Additional data file [5](#s5){ref-type="supplementary-material"}) and the corresponding figure legend (Additional data file [6](#s6){ref-type="supplementary-material"}). Supplementary Material ====================== ::: {.caption} ###### Additional data file 1 The processed raw data for the time course microarray study ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 2 Replicate 1 of the control microarray study with the ICI and CHX treatments alone ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 3 Replicate 2 of the control microarray study with the ICI and CHX treatments alone ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 4 The complete list of all 387 estrogen-responsive genes described in the article with UniGene cluster numbers from build 166 ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 5 Expression profiles of ICI and CHX responsive genes identified in the control experiments ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 6 The corresponding figure legend to expression profiles of ICI and CHX responsive genes identified in the control experiments ::: ::: {.caption} ###### Click here for additional data file ::: Acknowledgements ================ We thank Soek Ying Neo, Suk Woo Nam and Christopher Wong for assistance with the microarray technology, and Phillip Long, Joshy George and Radha Krishna Murthy Karuturi for helpful discussion on data analysis. We also thank George Reid for providing insightful comments on the manuscript. The research conducted at the Genome Institute of Singapore was supported by funding from the Biomedical Research Council (BMRC) of the Agency for Science, Technology, and Research (A\STAR) in Singapore. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Experimental design and microarray data selection of the time-course analysis of estrogen response in T-47D cells. (a)Cells were starved of serum and estrogen for 24 h before treatment with dimethysulfoxide (DMSO; carrier control), 17β-estradiol (E2), and E2 in combination with ICI 182,780 (ICI) and cycloheximide (CHX). Samples were taken at the 16 time points indicated. (b)Procedure for identifying direct ER targets. Data selection for estrogen-responsive genes was based on p-value cutoffs (p≤ 0.052) and magnitude of response (at least 1.2-fold change in the same direction for three time points). ICI sensitivity and CHX insensitivity were also determined by statistical measures, p≤ 0.058 and p≥ 0.24 respectively. Cutoff values were informed by expression profiles of cathespsin D and progesterone receptor, both known to be regulated by ER. ::: ![](gb-2004-5-9-r66-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Expression profiles of estrogen-responsive genes. (a)The 386 genes responsive to E2 were visualized by hierarchical clustering and the Eisen TreeView software. The columns represent time points arranged in chronological order, and each row represents the expression profile of a particular gene. By convention, upregulation is indicated by a red signal and downregulation by green. The magnitude of change is proportional to the brightness of the signal. Each panel represents a treatment condition as noted in the column headings. (b)ICI and (c)CHX treatments were included to identify the 137 ER-regulated genes and 89 primary response genes, respectively. ::: ![](gb-2004-5-9-r66-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Comparative analysis of estrogen-responsive gene expression profiles in T-47D and MCF-7 breast cancer cell lines revealed similar responses. (a)Expression data of 103 genes that were responsive to E2 and sensitive to ICI treatment in T47-D cells and present in the MCF-7 dataset \[31\] revealed concordant responses to E2 in 64% (66/103) of the genes (highlighted in magenta). (b)If the MCF-7 data is selected for E2-responsive and tamoxifen (Tam)-sensitive genes (see Materials and methods), there is a 24-gene overlap between the two datasets and the responses to E2 and the anti-estrogens ICI or tamoxifen are highly concordant (magenta) at nearly 96% (23/24). ::: ![](gb-2004-5-9-r66-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Estrogen-responsive genes identified in cell-line studies were also differentially expressed in ER^+^breast tumors compared to ER^-^tumors. Estrogen-responsive gene-expression profiles are compared to the composite expression ratio for those genes in each of the six breast tumor studies surveyed (L.D.M., B.M.F. Mow, L.A.V., and E.T.L., unpublished work and \[36-40\]). Differentially expressed genes were defined by p\< 0.01 between ER^+^and ER^-^tumor samples. Genes that responded similarly in ER^+^tumors and in vitrofollowing E2 treatment are highlighted in magenta. ::: ![](gb-2004-5-9-r66-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### ERE-like sequences are enriched in the extended promoter regions of putative target gene. (a)ERE predictions were made using a previously published model and optimized sensitivity setting. Prediction models were tested on the extended promoter regions (-3,000 to +500, both strands) from the most significant putative ER target and non-responsive genes, ordered by statistical significance. The y-axis represents the relative frequency of binding-site predictions as determined by the number of genes with predicted sites divided by the total number of genes. The number of most significant genes queried is indicated on the x-axis. Frequency of ERE predictions in putative target genes is significantly greater (p= 0.0027) than the similarly ranked non-responsive genes. Binding-site predictions were also carried out using position weight matrices describing sites for (b)Sp1, (c)AP-1 and (d)GATA1. Both Sp1 and AP-1 are known to be involved in regulating ER binding in certain target genes. GATA1 sites were included as negative controls. There is no significant enrichment of these sites in the putative target genes (see p-values in figure). ::: ![](gb-2004-5-9-r66-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Comparison of human and mouse orthologs. (a)Statistics from the comparative analysis of predicted EREs in human and mouse orthologous putative target gene pairs. (b)Venn diagram showing that out of the 72 orthologous pairs extracted for analysis, only 22 pairs have ERE predictions made in both the human and mouse sequences. ::: ![](gb-2004-5-9-r66-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### ER binds promoter regions encoding both conserved and non-conserved predicted response elements in an estrogen-dependent manner. (a)EREs (underlined) found upstream of NRIP1and GREB1coding regions are conserved in human, chimpanzee, mouse, and rat genomes. (b)PCR primers flanking the predicted conserved (NRIP1and GREB1) and non-conserved (ABCA3and TFF1/pS2) EREs were designed to detect ER binding following ChIP assays. The relative positions of the primers and ERE, relative to the TSS, are indicated. (c)Interactions between ER and predicted EREs were enhanced by estrogen treatment. MCF-7 cells were either mock-treated with the carrier dimethyl sulfoxide (-E2, gray bars) or treated with estradiol (+E2, black bars), followed by ChIP experiments. Black and gray bars indicate the enrichment of the binding site in anti-ER ChIP experiments over anti-GST ChIP experiments. Enrichment of all EREs was observed in hormone-treated cells whereas the mock-treated cells displayed less or very little enrichment. There was no enrichment of actin exon 3 control region or any of the input controls (open bars). ::: ![](gb-2004-5-9-r66-7) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### List of putative ER target genes ::: Accession number Symbol Gene name Number of time points\ Gene ontology ------------------------------------------------------------------------ ----------------- ------------------------------------------------------------------------------------ ------------------------- ----------------------------------------------------- (a) Putative ER target genes that are upregulated (59 out of 89) AL049265 mRNA; cDNA DKFZp564F053 14 Biological\_process unknown \[0000004\] NM\_003246 THBS1 thrombospondin 1 13 Cell adhesion \[0007155\] NM\_016339 Link-GEFII Link guanine nucleotide exchange factor II 13 Neurogenesis \[0007399\] U79299 OLFM1 Olfactomedin 1 13 Signal transduction \[0007165\] AK026062 DNAJC1 DnaJ (Hsp40) homolog, subfamily C, member 1 12 Protein folding \[0006457\] M62403 *IGFBP4^†^ Insulin-like growth factor binding protein 4 12 Signal transduction \[0007165\] NM\_001089 ABCA3^†^ ATP-binding cassette, sub-family A (ABC1), member 3 12 ATP-binding cassette (ABC) transporter \[0004009\] NM\_003714 STC2* Stanniocalcin 2 12 Cell-cell signaling \[0007267\] AF075060 Hypothetical protein DKFZp761P0423 11 Biological\_process unknown \[0000004\] AF271070 SLC38A1 Solute carrier family 38, member 1 11 Amino acid transport \[0006865\] AK024639 Hypothetical protein FLJ20986 11 Cation transport \[0006812\] AL080199 mRNA; cDNA DKFZp434E082 11 Biological\_process unknown \[0000004\] NM\_014365 H11 Protein kinase H11 11 Translational regulation, initiation \[0006446\] AF245389 *GREB1^†‡^ GREB1 protein 10 High-affinity zinc ion transport \[0006830\] NM\_002894 RBBP8^†^ Retinoblastoma binding protein 8 10 DNA repair \[0006281\] NM\_005067 SIAH2* Seven in absentia homolog 2 (Drosophila) 10 Ubiquitin-dependent protein degradation \[0006511\] U16752 *CXCL12^†^ Chemokine (C-X-C motif) ligand 12 (stromal cell-derived factor 1) 10 Immune response \[0006955\] AF086500 FZD8* Frizzled homolog 8 (Drosophila) 9 Biological\_process unknown \[0000004\] AF176012 JDP1 J domain containing protein 1 9 Physiological processes \[0007582\] AF182416 NIF3L1 NIF3 NGG1 interacting factor 3-like 1 (S. pombe) 9 DNA methylation \[0006306\] AK023772 Hypothetical protein FLJ13710 9 Developmental processes \[0007275\] AK024361 Hypothetical protein FLJ14299 9 Transcription regulation \[0006355\] AK025812 cDNA: FLJ22159 fis, clone HRC00251, mRNA sequence 9 Biological\_process unknown \[0000004\] NM\_001037 SCN1B Sodium channel, voltage-gated, type I, beta polypeptide 9 Sodium transport \[0006814\] NM\_003287 TPD52L1 Tumor protein D52-like 1 9 Signal transducer \[0004871\] NM\_003646 DGKZ Diacylglycerol kinase, zeta 104 kDa 9 Signal transduction \[0007165\] NM\_014333 IGSF4 Immunoglobulin superfamily, member 4 9 Virulence \[0009406\] NM\_016300 ARPP-21 Cyclic AMP-regulated phosphoprotein, 21 kD 9 Biological\_process unknown \[0000004\] AF200341 HPYR1 Helicobacter pyloriresponsive 1 8 Biological\_process unknown \[0000004\] AK023199 cDNA FLJ13137 fis, clone NT2RP3003150, mRNA sequence 8 Biological\_process unknown \[0000004\] AK025571 Hypothetical protein FLJ21918 8 RNA processing \[0006396\] D00265 CYCS Cytochrome c, somatic 8 Electron transport \[0006118\] NM\_000926 *PGR^†^ Progesterone receptor 8 Signal transduction \[0007165\] NM\_001634 AMD1* S-adenosylmethionine decarboxylase 1 8 Polyamine biosynthesis \[0006596\] NM\_002184 IL6ST^‡^ Interleukin 6 signal transducer (gp130, oncostatin M receptor) 8 Signal transduction \[0007165\] NM\_003489 *NRIP1^†^ Nuclear receptor interacting protein 1 8 Transcription regulation \[0006355\] NM\_004878 PTGES* Prostaglandin E synthase 8 Prostaglandin metabolism \[0006693\] NM\_012111 C14orf3 Chromosome 14 open reading frame 3 8 Protein folding \[0006457\] NM\_015878 OAZIN Ornithine decarboxylase antizyme inhibitor 8 Polyamine biosynthesis \[0006596\] AK023680 PPP1R15B Protein phosphatase 1, regulatory (inhibitor) subunit 15B 7 Biological\_process unknown \[0000004\] NM\_001909 *CTSD^†^ Cathepsin D (lysosomal aspartyl protease) 7 Proteolysis and peptidolysis \[0006508\] NM\_003774 GALNT4* UDP-N-acetyl-alpha-D-galactosamine:polypeptide N-acetylgalactosaminyltransferase 4 7 Biological\_process unknown \[0000004\] NM\_014810 CAP350 Centrosome-associated protein 350 7 Non-selective vesicle transport \[0006899\] NM\_016391 HSPC111 Hypothetical protein HSPC111 7 Leading strand elongation \[0006272\] AK021773 cDNA FLJ11711 fis, clone HEMBA1005152, mRNA sequence 6 Biological\_process unknown \[0000004\] AK025766 BRI3BP BRI3 binding protein 6 Biological\_process unknown \[0000004\] L23401 Human repeat region mRNA 6 Biological\_process unknown \[0000004\] NM\_000427 LOR Loricrin 6 Cell shape and cell size control \[0007148\] NM\_001932 MPP3 Membrane protein, palmitoylated 3 (MAGUK p55 subfamily member 3) 6 Signal transduction \[0007165\] NM\_002227 JAK1 Janus kinase 1 (a protein tyrosine kinase) 6 Protein phosphorylation \[0006468\] NM\_006392 NOL5A Nucleolar protein 5A (56 kDa with KKE/D repeat) 6 Transcription \[0006350\] NM\_006796 AFG3L2 AFG3 ATPase family gene 3-like 2 (yeast) 6 Biological\_process unknown \[0000004\] NM\_013324 CISH Cytokine inducible SH2-containing protein 6 JAK-STAT cascade \[0007259\] NM\_016233 PADI3 Peptidyl arginine deiminase, type III 6 Protein modification \[0006464\] NM\_020120 UGCGL1 UDP-glucose ceramide glucosyltransferase-like 1 6 Protein modification \[0006464\] AB037842 KIAA1421^‡^ KIAA1421 protein 5 RNA dependent DNA replication \[0006278\] NM\_001116 ADCY9 Adenylate cyclase 9 5 Signal transduction \[0007165\] NM\_014121 PRO0233 protein 5 Double-strand break repair \[0006303\] NM\_018053 Hypothetical protein FLJ10307 5 Cell death \[0008219\] (b) Putative ER target genes that are downregulated (30 out of 89) NM\_012342 *NMA^†^ Putative transmembrane protein 13 Melanin biosynthesis from tyrosine \[0006583\] AF039944 NDRG1* N-myc downstream regulated gene 1 11 Biological\_process unknown \[0000004\] AL049471 mRNA; cDNA DKFZp586N012 11 Biological\_process unknown \[0000004\] AK024964 NFIA Nuclear factor I/A 9 DNA replication \[0006260\] M16006 SERPINE1 Serine (or cysteine) proteinase inhibitor, clade E, member 1 9 Acute-phase response \[0006953\] NM\_004438 EPHA4 EphA4 9 Signal transduction \[0007165\] NM\_006449 CDC42EP3 CDC42 effector protein (Rho GTPase binding) 3 9 Signal transduction \[0007165\] AK026298 NMES1 Normal mucosa of esophagus specific 1 8 Biological\_process unknown \[0000004\] D16875 Human HepG2 3\' region cDNA, clone hmd1f06, mRNA sequence 8 Biological\_process unknown \[0000004\] NM\_002237 KCNG1 Potassium voltage-gated channel, subfamily G, member 1 8 Potassium transport \[0006813\] NM\_003032 SIAT1 Sialyltransferase 1 (beta-galactoside alpha-2,6-sialytransferase) 8 Protein modification \[0006464\] NM\_006605 RFPL2 Ret finger protein-like 2 8 Protein binding \[0005515\] AK025922 Hypothetical protein FLJ22269 7 Developmental processes \[0007275\] NM\_000504 F10 Coagulation factor X 7 Proteolysis and peptidolysis \[0006508\] NM\_001139 ALOX12B Arachidonate 12-lipoxygenase, 12R type 7 Epidermal differentiation \[0008544\] NM\_004354 CCNG2 Cyclin G2 7 Cell cycle checkpoint \[0000075\] NM\_006137 CD7 CD7 antigen (p41) 7 Humoral defense mechanism \[0006959\] NM\_007273 REA^‡^ Repressor of estrogen receptor activity 7 Negative control of cell proliferation \[0008285\] NM\_014583 LMCD1 LIM and cysteine-rich domains 1 7 Transcription factor \[0003700\] NM\_017572 MKNK2 MAP kinase-interacting serine/threonine kinase 2 7 Protein phosphorylation \[0006468\] D49356 Human mRNA (S100C-related gene) 6 Cell cycle \[0007049\] NM\_001878 CRABP2 Cellular retinoic acid binding protein 2 6 Signal transduction \[0007165\] NM\_004388 CTBS Chitobiase, di-N-acetyl- 6 Carbohydrate metabolism \[0005975\] NM\_006622 SNK serum-inducible kinase 6 Protein phosphorylation \[0006468\] AK022072 cDNA FLJ12010 fis, clone HEMBB1001635, mRNA sequence 5 Biological\_process unknown \[0000004\] AL137529 Hypothetical protein FLJ23751 5 Lipid metabolism \[0006629\] NM\_000430 PAFAH1B1 Platelet-activating factor acetylhydrolase, isoform Ib, alpha subunit 45 kDa 5 Signal transduction \[0007165\] NM\_013332 HIG2 Hypoxia-inducible protein 2 5 Biological\_process unknown \[0000004\] NM\_014770 CENTG1 Centaurin, gamma 1 5 Cell growth and/or maintenance \[0008151\] NM\_001719 *BMP7^†^ Bone morphogenetic protein 7 (osteogenic protein 1) 4 Cell growth and/or maintenance \[0008151\] \Number of time points that met the 1.2-fold change in the same direction selection criteria. ^†^Genes in bold have previously been shown to be direct targets in other ER^+^breast tumor cell lines (ZR75-1 and/or MCF-7). ^‡^CHX treatments alone had agonistic effects on these genes; therefore their CHX sensitivity in the presence of E2 is unclear. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparative analysis of predicted EREs in 22 human and mouse orthologous gene pairs ::: Gene symbol Human Mouse \% Identity\* Relative position\* (Δd) ------------- ------------ ------- --------------- ---------------------------- ----- ---- ------- ALOX12B NM\_001139 3 NM\_009659 4 100 87 40 NRIP1 NM\_003489 2 NM\_173440 4 100 72 70 GREB1 NM\_014668 2 NM\_015764 2 100 72 1,720 ACP33 NM\_016630 3 NM\_138584 4 100 66 46 F10 NM\_000504 2 NM\_007972 6 90 49 2,003 SERPINE1 NM\_000602 4 NM\_008871 3 90 23 41 CTSD NM\_001909 3 NM\_009983 1 90 26 1,310 OAZIN NM\_015878 1 NM\_018745 1 90 19 2,264 PADI3 NM\_016233 4 NM\_011060 2 90 21 4,153 Unknown NM\_017770 2 NM\_019423 1 90 28 3,828 JDP1 NM\_021800 1 NM\_013888 1 90 30 1,089 NIF3L1 NM\_021824 1 NM\_022988 2 90 28 955 SCN1B NM\_001037 2 NM\_011322 1 80 36 2,716 STC2 NM\_003714 1 NM\_011491 2 80 26 2,638 LOR NM\_000427 2 NM\_008508 2 70 32 4,253 ADCY9 NM\_001116 4 NM\_009624 1 70 28 3,042 AHCYL1 NM\_006621 1 NM\_145542 2 70 32 2,190 CISH NM\_013324 1 NM\_009895 3 70 19 1,624 HSPC111 NM\_016391 1 NM\_178605 1 70 19 1,302 FLJ22269 NM\_032219 2 NM\_172883 1 70 47 3,391 DNAJC1 NM\_022365 1 NM\_007869 1 60 30 1,672 CDC42EP3 NM\_006449 1 NM\_026514 1 50 25 451 \*The % identity and relative positions of EREs refer to the predicted pairs with the highest conservation between the two organisms. :::	PubMed Central	2024-06-05T03:55:48.505799	2004-8-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522873/", "journal": "Genome Biol. 2004 Aug 12; 5(9):R66", "authors": [ { "first": "Chin-Yo", "last": "Lin" }, { "first": "Anders", "last": "Ström" }, { "first": "Vinsensius Berlian", "last": "Vega" }, { "first": "Say", "last": "Li Kong" }, { "first": "Ai", "last": "Li Yeo" }, { "first": "Jane S", "last": "Thomsen" }, { "first": "Wan Ching", "last": "Chan" }, { "first": "Balraj", "last": "Doray" }, { "first": "Dhinoth K", "last": "Bangarusamy" }, { "first": "Adaikalavan", "last": "Ramasamy" }, { "first": "Liza A", "last": "Vergara" }, { "first": "Suisheng", "last": "Tang" }, { "first": "Allen", "last": "Chong" }, { "first": "Vladimir B", "last": "Bajic" }, { "first": "Lance D", "last": "Miller" }, { "first": "Jan-Åke", "last": "Gustafsson" }, { "first": "Edison T", "last": "Liu" } ] }
PMC522874	Background ========== The Caudata (tailed amphibians such as salamanders) are a major focus of work in vertebrate evolution and speciation \[[@B1],[@B2]\]. The salamander is also an important vertebrate model organism for understanding regeneration, being one of the few vertebrates that is able to regenerate entire body structures such as the limb, tail and jaw as an adult. Despite the pivotal role of this animal order in research, comparatively little sequence information is available. In contrast, 458,413 nucleotide sequences exist for the Anura (frogs and toads). This high number is primarily attributable to large EST sequencing efforts for the model organisms for embryology - Xenopus laevisand Silurana tropicalis. A salamander EST project is particularly important as these organisms have extremely large genomes, making a genome project unwieldy and unlikely without specialized approaches such as methylation filtration \[[@B3]\]. Genome sizes range from 8.5 billion base pairs for Desmognathus monticola(seal salamander) to nearly 70 billion base pairs for Plethodon vandykei(Van Dyke\'s salamander) \[[@B4]\]. The ambystomatid Ambystoma mexicanum, a species important for studies in evolution, regeneration and development, has an estimated genome size between 21.9 billion and 48 billion base pairs \[[@B5],[@B6]\] and measurements of its genome in centimorgans (cM) has yielded the largest size reported for a living vertebrate so far (7,291 cM \[[@B7]\]). In maize, another organism with a large genome, 60,000 sequence reads were required before genome sequencing of methylation-filtered genomic libraries generated significantly more gene sequence information than the available maize EST sequences \[[@B8]\]. Molecular evolution studies of salamanders have relied primarily on mitochondrial genes such as those for ribosomal RNAs and cytochrome c\[[@B9]\]. The lack of sequence information among the Caudata hinders the ability to perform sequence comparison with other important gene families. Furthermore, because of the lack of clones, the number of molecular markers available to study salamander embryology and regeneration is low. To address this gap in sequence availability we have generated a large gene sequence set for A. mexicanum. We chose this species because of its role in evolutionary, developmental and regeneration studies. A. mexicanumis easily bred in the laboratory, and animals can be obtained from a large, NSF-funded colony \[[@B10]\]. We have sequenced inserts from two cDNA libraries, one produced from dorsal regions of stage 18-22 embryos, consisting primarily of neural tube, somite and notochord. The second library was constructed from day-6 regenerating tail blastema tissue. By sequencing from these two sources, our goal was to obtain sequences of transcripts involved in organizing and regenerating the primary body axis. Here we describe the EST gene set, provide an example of molecular phylogenetic analysis of one gene from this collection, and describe the database created for organizing the A. mexicanumEST information. This database is also being implemented for EST sequences from a full-length X. laeviscDNA library, and for sequences from a Canis familiarisEST project. Results ======= Assessment of library and EST sequence quality ---------------------------------------------- To generate a diverse set of sequences involved in organizing and regenerating the primary body axis, two independent cDNA libraries were used for sequencing. One was derived from dorsal regions of stage 18-22 embryos containing neural tube, somite and notochord - called the \'neural tube\' library - the other from 6-day post-amputation regenerating tail blastema. From 18,432 sequencing attempts 17,522 high-quality sequences were obtained after Phred analysis \[[@B11]\]. All sequences are 5\' reads of the inserts. Of 17,522 high-quality, single-pass sequencing runs, 32 clones contained no insert and 137 sequences were below 32 base pairs (bp). These sequences were excluded from further analysis (32 bp representing the lower limit for assembly of a sequence using TIGR-assembler), yielding 17,352 clones for final analysis. The neural tube library was the origin of 7,469 sequences and the blastema library of 9,883 sequences (Table [1](#T1){ref-type="table"}, and see Materials and methods). As shown in Figure [1a](#F1){ref-type="fig"}, the average sequence read length peaked between 500 and 600 nucleotides with an average length of 510 nucleotides and a maximum of 871. The blastema and neural tube libraries were unnormalized and unamplified. We assessed library quality and diversity on the basis of the number of redundant clones in the library. Redundancy was estimated by performing BLASTN searches \[[@B12]\] against all clones sequenced. After sequencing 10,752 clones of the blastema library 42% of the sequences were still unique, and 50% of clones were still singlets after sequencing 7,680 clones from neural tube, indicating that both libraries display high diversity. EST assembly into contigs ------------------------- To identify ESTs belonging to the same open reading frames (ORFs), sequences were assembled into contigs using TIGR-Assembler version 2 \[[@B13]\]. The 17,353 sequences assembled into 6,594 contigs, of which 217 were less than 100 nucleotides long and excluded from further analysis. A total of 6,377 contigs was therefore left for final analysis (Table [1](#T1){ref-type="table"}). Of these, 4,561 contigs contained a single clone. The average contig length of the remaining dataset was 616 nucleotides (Figure [1b](#F1){ref-type="fig"}). Other than singlets, most of the contigs consisted of two ESTs (884 contigs, Figure [1c](#F1){ref-type="fig"}). The largest contigs included cytochrome coxidase subunit I (469 ESTs), 12S rRNA (445 ESTs), nuclear factor 7 Zn-binding protein A33 (332 ESTs), type II keratin (274 ESTs), keratin (211 ESTs) and cytoplasmic beta-actin (206 ESTs) (Table [2](#T2){ref-type="table"}). Comparison to existing A. mexicanumgenes in NCBI: 6,000 new contig sequences ------------------------------------------------------------------------------ A total of 1,134 ESTs were available from A. mexicanumin the National Center for Biological Information (NCBI) EST databases prior to this work, most of which originate from a sequencing effort of the Voss laboratory (\[[@B14]\] and S.R. Voss, D. King, N. Maness, J.J. Smith, M. Rondet, S.V. Bryant, D.M. Gardiner, and D.M. Parichy, unpublished work (NCBI-accession numbers BI817205-BI818091); see also \[[@B15]\]). We examined to what extent our EST dataset overlapped with the sequences available to date. Only 600 of the ESTs in the public database identified one of our contigs in a BLASTN search as a homolog; in 85% of cases, the E-value was below 1E-50 and the sequences can be considered as potentially identical. Existing ESTs in the database largely originate from regenerating limb (S.R. Voss, D. King, N. Maness, J.J. Smith, M. Rondet, S.V. Bryant, D.M. Gardiner and D.M. Parichy, unpublished work). There was, however, only a slight bias of matching contigs to regenerating blastema (49%) as compared to neural tube (44%). Seven percent of identified contigs were found in both libraries. These results mean that our EST data enriches the existing sequence resource of A. mexicanumwith approximately 6,000 new gene sequences. BLAST analysis of A. mexicanumcontigs to assign homologies ------------------------------------------------------------ To identify putative homologies to known proteins, we subjected the contigs to BLASTX searches against the non-redundant protein database (NR, NCBI) where a cutoff E-value of 1e-05 was used for parsing output files. In our annotation, we used an E-value of 1e-20 as an upper limit to assign significant homology. We note that this does not imply that such sequences are true orthologs. In addition, in cases where no significant homology was found, we used an E-value limit of 1e-05 to designate weak homology. We find this additional category of \'weak homology\' useful for data mining. As most contigs do not represent full-length sequences, it is possible that only a highly divergent region of a gene sequence is available in our collection. The category of weak homology allows us to find potential homologs in such situations. For example, the BLAST search for contig Am\_4671 yielded the GenBank entry NP\_004055, cyclin-dependent kinase inhibitor 1B (Homo sapiens), as the top hit with an E-value of 4e-07. This assignment was based on the carboxy-terminal 120 amino acids of the protein, which represents the less conserved region. When we isolated a full-length clone for Am\_4671 from our library, we could confirm that it is indeed the axolotl ortholog of cyclin-dependent kinase inhibitor 1B (p27^Kip1^), as discussed later. Taken together, a total of 3,718 (58%) sequences shared homology with a protein from selected model organisms in the non-redundant database and could be assigned a putative identity. The E-value distribution of the top hits in the non-redundant database is shown in Figure [2a](#F2){ref-type="fig"}. Of the contigs, 11% matched a protein with an E-value below 1e-99 and are therefore likely to be true orthologs. Seventy percent of the contigs found a hit with an E-value between 1e-20 and 1e-99 and were assigned significant homology. Finally, 19% of contigs had a first hit with an E-value between 1e-19 and 1e-05 and were assigned weak homology to a protein from the non-redundant database. For annotating our database, these top hits from human, mouse (Mus musculus), rat (Rattus norvegicus), frog (X. laevis), zebrafish (Danio rerio), fugu (Takifugu rubripes), fruitfly (Drosophila melanogaster), mosquito (Anopheles gambiae), worm (Caenorhabditis elegans), newts and the yeast species Saccharomyces cerevisiae, Schizosaccharomyces pombeand Candida albicanswere collected and the closest homolog from the above species was used to assign a putative identity. To estimate how many of the clones are full length we examined the BLAST alignments for the position of the alignment in respect to the database sequence. Of the 3,718 sequences with homologs, 1,107 (29.8%) could be aligned in the amino terminus (with the alignment starting before position 10). As the library was poly(dT) primed, many of these clones are likely to represent full-length inserts. Of these 199 (5.4%) could be aligned from the amino terminus to the carboxy terminus and are potential full-length sequences. Forty percent of our EST sequences did not generate a significant hit in the non-redundant protein database. The availability of additional sequence databases including complete genome sequences from several organisms allowed us to expand our BLAST searches to identify all possible homologs to the A. mexicanumcontigs. With the remaining set of contigs, we first performed BLASTN searches against the nucleotide non-redundant (NT) database and BLASTX searches against the EST database. Finally, we performed BLASTX searches against the fugu and human proteomes. In all cases, an E-value of 1e-05 was used to assign potentially homologous sequences. Sequences in the NT database identified an additional 134 contigs and a further 220 contigs found a hit in the EST databases. A homolog was found for 3,340 (52%) contigs in the fugu proteome and 3,698 (58%) contigs shared homology with a protein from the human proteome. In total, an additional 468 contigs identified a homolog in the selected databases beyond the original assignment from the non-redundant protein database (Figure [2b](#F2){ref-type="fig"}). Gene sequences with no identifiable homology -------------------------------------------- No homologous sequence could be found for 2,191 (34%) contigs in any of the databases searched. Because the library was poly(dT) primed, many of these sequences could represent 3\' untranslated regions (3\' UTRs). We determined that 953 sequences (43% of non-homologous contigs) contained no ORF and were therefore potential untranslated regions. Thirty of the sequences shared homology to an existing A. mexicanumclone from the EST database (Table [3](#T3){ref-type="table"}). The complete list of unique ESTs can be downloaded from \[[@B16]\]. Assignment of the A. mexicanumdataset to common Gene Ontology terms --------------------------------------------------------------------- From the homologous proteins found, contigs were assigned a biological process, molecular function and cellular component from the Gene Ontology (GO) database \[[@B17]\]. The closest annotated homolog in the GO database was used, using an E-value of 1e-20 as a cutoff, for assigning these categories. A biological process could be assigned to 2,156 contigs (34% of all contigs and 58% of those sharing a homolog in the non-redundant database); 2,186 contigs (34% and 59%, respectively) were assigned a molecular function; and 2,198 contigs (34% and 59%, respectively) could be assigned a cellular component. The most abundant molecular function assigned was \'death receptor interacting protein\', followed by \'peptidase\', the highest-ranking biological process were \'biological process unknown\' and \'proteolysis/peptidolysis\' and the most abundant cellular components assigned were the \'actin cytoskeleton\' and \'transcriptional repressor complex\'. The largest fraction of the contigs was assigned a cellular process in the GO category biological process (87% of annotated contigs) (Figure [3a](#F3){ref-type="fig"}). We split the biological processes further into different categories: the most abundant categories were \'protein metabolism/modification\' (18% of assigned contigs); \'housekeeping functions/metabolism\' (17%); \'intracellular transport\' (15%); \'cell cycle/proliferation\' (13%); \'RNA metabolism\' (13%); \'intracellular signaling\' (8%); and \'DNA metabolism/repair\' (5%) (Figure [3a](#F3){ref-type="fig"}, Table [4](#T4){ref-type="table"}). A list of annotated contigs is downloadable from \[[@B16]\]. Common SMART and PFAM domains in the A. mexicanumdataset ---------------------------------------------------------- To identify potential domains in the axolotl contigs, we performed RPS-BLAST searches against the conserved domain database (CDD, NCBI) \[[@B12],[@B18]\] using the default cutoff E-value of 0.01. A total of 2,199 (34.5%) contigs had a known protein domain in either the CDD or the SMART or PFAM databases. A detailed list of common protein domains identified in our dataset is given in Table [5](#T5){ref-type="table"}. Among the protein domains identified were homeobox domains such as HOX, PAX and Prox1, eight helix-loop-helix (HLH) domains, RNA-binding domains such as KH and RRM, 69 kinase domains, metal- and lipid binding domains and domains involved in cell-cycle control and ubiquitination (RING fingers, HECT domains, three cullin domains and 12 cyclin domains). Many of these domains were annotated for the first time in a sequence from A. mexicanum. We also compared the occurrence of those domains in other vertebrate species. For most of the common protein domains, only a fraction were found in our dataset; many of these are quite abundant compared to X. laevisor Gallus gallus. The RNA-binding domains KH and RRM especially showed high abundance in our contigs. A complete list of domains is downloadable from \[[@B16]\]. We assigned cellular functions to the identified domains and analyzed the output according to the functional distribution of contigs (Figure [3b](#F3){ref-type="fig"}). The most abundant domains were found in the category \'intracellular transport\'; this is due to redundant annotations of small GTPases. The second largest fraction belonged to \'RNA-binding and metabolism\', followed by \'DNA-binding and transcriptional control\'. In silicodifferential display of A. mexicanumcontigs in blastema and neural tube ------------------------------------------------------------------------------------ ### Regeneration versus development We were interested to see if there were strong differences in the sequence representation of the libraries that reflect the different biological processes taking place in each tissue. To this end, we compared the representation of ESTs in the two libraries. This type of in silicodifferential display has been performed for ESTs in the NCBI collection, and, as with the NCBI differential display data, we have assessed the statistical significance of the differences using Fisher\'s exact test. A total of 104 contigs met the cutoff value of 0.005 in Fisher\'s exact test and can therefore be considered differentially expressed. Table [4](#T4){ref-type="table"} provides a detailed comparison of EST representation categorized according to their biological process annotation. Considering the biological properties of the blastema tissue versus the neural tube tissue, we were particularly interested in differential display results of gene sequences that had been assigned to the biological functions of RNA metabolism (as an indicator of an high proliferation index), cell cycle and proliferation and differentiation. The blastema library was produced from tail tissue that was in the process of forming the blastema progenitor cells for regeneration. Blastema formation involves dedifferentiation of mature cells, and entry into rapid cell cycles. In contrast, the neural tube library contains tissue undergoing cell specification and differentiation, such as neurogenesis and somitogenesis. Although these embryonic tissues are still proliferating, the proliferation index of the cells from neural tube should be lower than from blastema. ### RNA metabolism A total of 168 contigs annotated under RNA metabolism (127 when normalized to the ratio of sequenced ESTs from blastema and neural tube) were more frequently sequenced or uniquely sequenced in blastema (6% of assigned contigs, 2.6% of all contigs). This group included RNA metabolism, RNA processing, splicing, editing, nuclear export, binding, catabolism, cleavage, capping, rRNA modification, rRNA transcription and tRNA aminoacetylation. Forty-five contigs assigned a process in RNA metabolism were upregulated or unique in neural tube (2% of assigned and 0.7% of all contigs). After Fisher\'s exact test analysis, 24 of the clones were considered differentially regulated in the two libraries; 22 out of the 24 contigs were enriched or unique in blastema (Table [4](#T4){ref-type="table"}). ### Cell cycle and proliferation 126 contigs (95 when normalized to sequencing ratios) were assigned as cell-cycle genes (5% of assigned contigs and 1.5% of total contigs) and were more frequently sequenced or uniquely sequenced in the blastema library, compared with 52 in the neural tube library (2.5% and 0.8%, respectively). This category included regulation of mitosis, mitosis, cell-cycle regulation, regulation of cyclin-dependent kinase (CDK) activity, cell proliferation, DNA replication, M phase, mitotic spindle checkpoint, mitotic spindle assembly, chromosome segregation and cytokinesis. As an example, 10 different types of cyclins were found, from various stages of the cell cycle. Seven of the contigs found in cell-cycle regulation met the cutoff criteria of statistical significance in Fisher\'s exact test. Five out of the seven contigs were more highly represented or unique in blastema (Table [4](#T4){ref-type="table"}). ### Differentiation Whereas proliferation-associated genes were found with a higher sequence representation in the blastema library, genes that had been electronically annotated as involved in \'cell differentiation\' had a higher representation in the neural tube library. A total of 28 contigs were electronically assigned the biological process \'differentiation\'. After Fisher\'s exact test, five contigs showed differential regulation in this group. Three out of the five contigs were found in neural tube (Table [4](#T4){ref-type="table"}). Taken together, these results indicate that the two cDNA libraries have differences in sequence representation that appear to correlate with the physiological processes taking place in the two tissues. Gene families involved in cell-cycle control and development in the A. mexicanumdataset ----------------------------------------------------------------------------------------- As mentioned earlier, the Mexican axolotl is an important model organism for a number of reasons. First, it is the premier vertebrate model for studying regeneration. Second some aspects of caudate development, for instance mesoderm involution and notochord formation, more closely resemble those found in higher vertebrates than do those in other amphibian embryological models such as X. laevis\[[@B19]\]. Finally, the axolotl has interesting developmental features, particularly in relation to metamorphosis. The axolotl undergoes \'cryptic metamorphosis\', which is defined by its existence in a perrenibranchiate state and retaining some larval features into adulthood (for instance gills, larval skin morphology, caudal fins). The animals become sexually mature in this state, and develop only small rudimentary lungs. So far, very few markers are available to study these processes in this organism. We examined our dataset for genes that are potentially useful for studying regeneration features or developmental processes. To this end, we analyzed our data for genes that are either involved in regulating the cell cycle - as would be expected for the highly proliferative tissue of a regenerating body structure - or could play an essential role during development and metamorphosis from the larval to the adult stage. A list of genes that could be assigned to either cell-cycle regulation or development is shown in Table [6](#T6){ref-type="table"}. Among the genes involved in cell-cycle regulation were A-, B- and E-type cyclins, cyclin-dependent kinase 4 (Cdk4), Polo kinase, the kinase inhibitor p27^Kip1^, the protein phosphatase Cdc25A, as well as the anaphase-promoting complex (APC) activator proteins Cdc20 and Cdh1. Representing genes involved in developmental processes, we found transcription factors such as HoxA2, B12, C4 and C8, Pax6, as well as Cdx1 and Cdx2. Furthermore we found several genes for proteins that are part of the transforming growth factor-beta (TGF-β) signaling pathway, such as TGF-β, bone morphogenetic protein 1 (BMP-1), BMP and activin membrane-bound inhibitor, activin receptor type II, as well as the transcription factors Smad5 and Smad8. Genes for proteins such as Smad8 and BMPs might be of especial interest to the research field of embryonic development, as they have been associated with mesoderm involution \[[@B20]\]. Other important developmental genes that could be found in our dataset include those for Wnt5 and Wnt8, Sonic hedgehog, retinoblastoma binding protein 2, beta-catenin, as well as Frizzled 2, 5 and 7. Finally, it has been shown that the thyroid hormone receptor pathway has an essential role in the timing of metamorphosis in A. mexicanum\[[@B21]-[@B23]\]. We identified the protein TRIP12 (thyroid hormone receptor interacting protein 12), which is a HECT-domain-containing ubiquitin ligase and could have an essential role in regulating thyroid hormone response during development and/or metamorphosis. Phylogenetic analysis of the CDKN1 gene family in vertebrates: amphibians contain an unusual CDKN1 family member ---------------------------------------------------------------------------------------------------------------- The EST collection will provide rich data for the phylogenetic comparison of particular genes. Cell cycle and cell differentiation are cellular functions that have been modified in various organisms through evolution and it will be interesting to understand the evolutionary basis of such changes. Here we analyze a particularly interesting gene family, the CDKN1 family of cell-cycle regulators which inhibit cell-cycle progression by binding to and inactivating CDKs. As a starting point for phylogenetic analysis, the mitochondrial 12S ribosomal RNA gene from our collection resulted in the expected tree, with the anuran amphibian X. laevisand the caudate A. mexicanumgrouping together compared to other vertebrates such as fish, birds and mammals (Figure [4a](#F4){ref-type="fig"}). Next, we constructed an unrooted phylogenetic tree to compare members of the cyclin B family - cyclins B1, B2 and B3. The sequences of each family member formed strictly separate groups, with the A. mexicanumand X. laeviscyclin B1, B2 and B3 genes grouping with their vertebrate orthologs (Figure [4b](#F4){ref-type="fig"}). In contrast, we obtained a quite different picture when we examined the CDKN1 family. In most vertebrates, this family consists of three members: p21 (CDKN1A), p27^Kip1^(CDKN1B) and p57 (CDKN1C). In X. laevis, however, only a single family member called p28^Kix1^(also called p27^Xic1^), which shows unusual sequence features compared to the p27 sequences from any other vertebrate species, had been described in the literature \[[@B24],[@B25]\]. We wondered whether A. mexicanumharbored the \'canonical\' p27^Kip1^or a p28^Kix1^similar to that of Xenopus. We initially searched our A. mexicanumdata for CDKN1 orthologs and, in contrast to Xenopus, we found a bona fidep27^Kip1^sequence that clusters closer to vertebrate p27^Kip1^sequences compared to the Xenopusp28^Kix1^(Figure [4c,d](#F4){ref-type="fig"}). Considering this interesting finding, we then undertook a more complete analysis of the CDKN1 family in vertebrates by searching for CDKN1 family members in several databases: the sequenced genomes from human, mouse, rat, fugu or zebrafish, the recently released genome sequence of X. tropicalis, the X. laevisEST collection, the zebrafish and fugu genomes, and a complementary A. mexicanumand A. tigrinumEST set generated by Putta et al. \[[@B26]\]. This data mining revealed two striking features about the distribution of CDKN1 family members among vertebrates (Table [7](#T7){ref-type="table"}). First, the p28^Kix1^orthologs were only found in amphibians (X. tropicalis, X. laevis, A. mexicanum, A. tigrinum tigrinum). We were not able to identify a p28^Kix1^-like gene in any other database. These p28 orthologs group as a distinct branch in an unrooted phylogenetic tree (Figure [4c,d](#F4){ref-type="fig"}). These data so far suggest that the p28 family is a CDK inhibitor that is specific for amphibians. With new genome sequence data being released, it will be interesting to see whether the most closely related lineage of birds contains a p28-like gene or whether this gene family is found solely in amphibians. Second, CDKN1B (p27^Kip1^) and CDKN1C (p57) were present in the A. mexicanumdatabases but were not found in either X. laevisor X. tropicalis, which have far more EST and genome sequence information (Table [7](#T7){ref-type="table"}, Figure [4c,d](#F4){ref-type="fig"}). While it is not possible to conclude definitively that Xenopusspecies lack these genes, the current data are highly suggestive of such a scenario. We examined in depth the phylogenetic relationships of the CDKN1 family members among vertebrates by constructing unrooted phylogenetic trees, either using the most conserved, amino-terminal 88-amino-acid domain, which includes the functionally important Cdk2-interaction region, or the entire coding sequence. Analysis of the amino terminus showed that while A. mexicanump27 and p57 clearly grouped with their respective orthologs from other vertebrates, the p28^Kix1^proteins from axolotl and the two Xenopusspecies clustered as a group distinct from any of the other CDKN1 families (Figure [4c](#F4){ref-type="fig"}). The p28^Kix1^family showed a closer relationship to p57 than to other CDKN1 members, branching off close to the p57 family. Phylogenetic analysis using the entire coding sequence of the CDKN1 genes, which includes the Cdk2- and PCNA-binding site, resulted in a closer grouping of p28 with the p27 branch (Figure [4d](#F4){ref-type="fig"}). In both cases, however, the p28 family clearly formed a separate group from the other CDKN1 families. The Ambystoma mexicanumEST database ------------------------------------- A relational database with a web-based front end was created to store, navigate and annotate analyzed contigs. The main object of the database is the annotated sequence contig, which contains information about its length, putative identity, computationally calculated expression profile, GO annotation, homologous proteins and identified domains, as well as number and identity of ESTs that build the contig (Figure [5a](#F5){ref-type="fig"}). The Gene Identifier (GI) and GO annotation can be modified by the administrator. To circumvent the problem of split contigs, we introduced a super-contig, to which related contigs can be assigned. Furthermore, the administrator can modify the relationship of EST to contig manually. All protein and domain alignments, as well as the assembly of the EST sequences of a contig are stored and can be viewed by the user. On the contig main page, three homologs at most from selected species are shown, with a full list of homologs from selected species displayed on the protein information page (Figure [5c](#F5){ref-type="fig"}). To make use easier, an image of the identified domains with the beginning and end base pair of the alignment is shown on the contig page. Individual ESTs can be accessed via the contig page, including their length, storage information, quality information and available trimmed EST-sequence (Figure [5b](#F5){ref-type="fig"}). Some of the main advantages of this database are: first, the direct links to source databases such as the NCBI sequence database, GO database, CDD, and the Smart and Pfam databases for identified domains; second, direct visualization of source data such as sequence alignments of contigs to homologs and domains, as well as alignments of EST assemblies; third, easy retrieval of sequences for further analysis like BLAST-searching; fourth, user-specific annotation of contigs; and fifth, easy manipulation and editing of contig annotations. The database will be available from \[[@B27]\]. Discussion ========== The salamander, and in particular the species A. mexicanum,represents an important vertebrate organism for evolutionary, developmental and regeneration studies. The salamanders provide an essential amphibian counterpoint to the anurans such as X. laevis, displaying distinct embryology and other physiological features. For example, mesoderm involution during gastrulation and subsequent notochord formation is distinctive between A. mexicanumand X. laevis. The characteristics of mesoderm involution in A. mexicanummore closely resemble those found in other vertebrates \[[@B19]\]. This and other evidence indicates that A. mexicanumand other urodele amphibians are likely to have retained more ancestral features in common with the \'primitive\' tetrapod compared to X. laevis, which appears to be more derived. It is interesting that we observed such segregation on the sequence level of the CDKN1 family. X. laevisappears to have a highly unusual make-up of CDKN1 family members. So far, CDKN1A (p21) and the highly derived p28^Kix1^are the only CDKN1 family members found in both X. laevisand X. tropicalis. In contrast, the ambystomatids appear to have all the members of the CDKN1-family - including p28^Kix1^- assuming that the p21 gene is missing purely as a result of lack of sequence information. In addition, our data suggest that p28 is an amphibian-specific variant of the CDKN1 family. Two major questions arise from these data: first, does the amphibian-specific p28 fulfill a cellular function that is unique to this phylogenetic lineage; and second, does the genotypic difference in the gene set of the CDKN1 family in the two amphibian species account for the macroscopic differences observed in developmental mechanisms. The fact that the CDKN1 family is an essential regulator of the cell cycle opens new possibilities for experimental research along these lines. Given the estimates in the number of genes present in the human genome (20,000-50,000) \[[@B28]\], we estimate that our EST contig set (6,377) contains between 10 to 25% of the total number of genes in the axolotl. While the database is not yet complete, it represents a significant proportion of the axolotl transcriptome. Further sequencing efforts, including an NIH-funded EST sequencing project for the axolotl \[[@B26]\], will enlarge the current dataset to provide a comprehensive gene sequence resource for this organism. Our analysis indicates that the majority of A. mexicanumgenes are homologous to genes present in other vertebrates. Sixty-six percent of contigs gave a significant match in either the non-redundant protein or nucleotide databases, the EST databases or the human and fugu protein databases. Thirty-four percent of contigs could not be assigned a homolog in any of the searched databases, and 44% of those could not be assigned a coding sequence and are therefore considered to be part of the UTR. Nineteen percent of the contigs seem to represent novel genes that have not been found in any other organism so far. The expressed sequence tags generated in this study also provide a large source of sequence information for developmental and regeneration studies. For example, an examination of the database yielded 194 genes involved in cell proliferation, including pivotal cell-cycle genes such as those for Cdc2, 10 different cyclin family members, Cdk4 and p27. A search for developmental molecules involved in intercellular communication yielded Wnt8, Wnt5B, FGF receptor 4a (FGFR4a), Sonic hedgehog, BMP receptor (BMPR) and BMP-1, while a search for homeodomain-containing proteins yielded 11 members, including Cdx1, Cdx2, HoxA2, HoxC8 and HoxB13. The ESTs were derived from two cDNA libraries, stage 18-22 embryonic neural tube/notochord/somite tissue, and day-6 regenerating tail tissue. The embryonic library represents a developmental stage where tissue specification is occurring, whereas the blastema library represents a tissue that is undergoing dedifferentiation, rapid proliferation and cell respecification. Accordingly, we find differences in transcript representation in the two libraries. The blastema library is particularly enriched in cell-cycle genes and RNA metabolism genes, presumably reflecting the high proliferative index of the early regenerating blastema. Conclusions =========== This set of 17,352 ESTs from A. mexicanumwas generated to provide a comprehensive sequence dataset for the community of biologists. Forty percent of genes could still be found in singlets, which reflects a high diversity of sequences in our cDNA set. Annotation of the assembled contigs revealed a substantial difference in gene representation in the two sequence libraries, reflecting their biological source - regenerating blastema being in a highly proliferative state and embryonic neural tube being a tissue undergoing differentiation. Sequence analysis of assembled contigs revealed that 64% of genes had a putative homolog in other species; 19.4% of the contigs contained a putative coding sequence and can be considered novel genes. From this, we conclude that A. mexicanumdoes not contain an unusually high number of organism-specific genes. The CDK inhibitor family CDKN1 was selected for comparative phylogenetic analysis. Unlike the frogs X. laevisand X. tropicalis, ambystomatids most probably contain all members of the CDKN1 family, including the amphibian-specific protein p28^Kix1^/p27^Xic1^, which shows unusual sequence divergence compared to CDKN1 members in other vertebrate species. Such data would support the contention that A. mexicanumis closer to a basal tetrapod compared to X. laevis. The EST sequences and annotated contigs presented in this paper will be a publicly available and useful resource for research in various fields. Materials and methods ===================== Plasmid cDNA library construction --------------------------------- Total RNA was purified using Trizol (Invitrogen) from 6-day regenerating tail blastemas and from neural tube-somite-notochord-containing tissue dissected from stage 18-22 A. mexicanumembryos. Total RNA quality was assessed by determining the relative brightness of the 28S:18S rRNA bands (2:1). For library construction mRNA was purified and size fractionated, then poly(dT)-primed cDNA was synthesized and directionally cloned into the NotI-SalI sites of the pCMVSport6 vector. DNA was transformed by electroporation into EMDH10B-TONA bacteria (library construction performed by Invitrogen). Two separate, unnormalized libraries were produced. The blastema library contained an average insert size of 1.67 kb and 2.67 × 10^7^independent transformants and the neural tube library had an average insert size of 1.5 kb and 1.9 × 10^7^transformants. From each library 100,000 clones were arrayed into 384-well plates (Resource Zentrum/Primary Database, Berlin, Germany). Sequencing ---------- For sequencing, single-pass reads from the 5\' end of the library inserts were performed using a custom-designed SP6 primer: GCACATTAGGCCTATTTAGGTGACA. DNA from bacterial library clones was amplified using the Templiphi reaction, based on φ29 rolling-circle replication of DNA (AP Biotech). Briefly, approximately 0.5 μl of bacterial glycerol stocks were picked up using 96-pin plastic replicators (Genetix) and centrifuged into 96-well PCR plates. Five microliters of denaturing buffer was added, and samples heated to 95°C for 3 min. After cooling, 5 μl Templiphi enzyme was added and samples incubated overnight in a 30°C incubator. The Templiphi reaction provides two advantages for large-scale sequencing projects on capillary sequencers. First, the reaction proceeds to an endpoint where all nucleotide is incorporated, yielding uniform quantities of DNA from varying amounts of starting bacteria (or DNA). Second, the rolling-circle reaction results in large pieces of DNA that, in contrast to plasmid DNA, do not enter the capillary and interfere with the sequencing run. For sequencing reactions, the DNA preparation was diluted fivefold with distilled water. Sequencing reactions were performed using the DYEnamic ET Dye terminator kit diluted twofold with DYEnamic ET dilution buffer (AP Biotech). Five microliters of DNA was added to 5 μl of sequencing reaction mix with primer and cycled 30 times under the following conditions: 95°C 20 sec, 60°C 1 min. Sequencing was performed on a MegaBACE 1000 (AP Biotech). Runs were either performed at injection: 3 kV 60 sec, run: 8 kV 120 min, or injection: 3 kV 60 sec injection, run: 3 kV 360 min. Analysis of library quality --------------------------- The redundancy of the arrayed libraries was tested by performing BLASTN searches \[[@B12]\] against all sequenced ESTs from the two libraries. Hits against clones other than the query with an E-value lower than 1e-50 were considered for clustering. Submission of ESTs to NCBI GenBank ---------------------------------- The sequences were submitted to GenBank. After quality control, individual ESTs were used to search the non-redundant protein database (release of July 2004) using the program BLASTX from the standalone NCBI-BLAST package \[[@B12]\]. For annotation of sequenced ESTs, the top hit of the BLAST output was used, whereby an E-value of 1e-20 was used for significant similarity and an E-value of 1e-05 was used as a cutoff value for weak similarity. Analysis and assembly of sequence data -------------------------------------- Quality control of sequenced ESTs was performed using the program Phred \[[@B11]\] using a cutoff of 20 for trimming low-quality regions, and vector trimming was performed using the program cross-match \[[@B11]\]. (We note here that the arbitrary Phred score reflects the likelihood of a false base. A Phred score of 20 indicates that in 1 out of 100 trials (10^2^), the base would be false, 30 would reflect a wrongly sequenced base in 1 of 1,000 trials (10^3^), and so forth.) Sequence and contig files can be downloaded at \[[@B16]\]. The resulting high-quality sequences were assembled into sequence contigs with the program TIGR-Assembler version 2 \[[@B13]\]. Alignment of contigs was performed with the program ClustalW with the settings Gap Opening 5 and Gap Extension 85 \[[@B29]\] or Cap3 \[[@B30]\], when ClustalW could not correctly assemble the sequences. Assembled contigs were used to perform BLAST searches (BLASTX, BLASTN from NCBI-BLAST \[[@B12]\]) against the non-redundant protein sequence database (release of November 2003), human and fugu protein databases and the NCBI EST database, all downloaded from the NCBI. Domain searches were done with RPS-BLAST against the conserved domain database (CDD \[[@B18]\]) from the NCBI. BLAST and domain-search output files were parsed for homologous sequences, whereby an E-value of 1e-05 was used as a cutoff for BLASTN and BLASTX searches against the sequence databases and the default cut-off of 0.01 was considered to yield significant homology to conserved domains from CDD. A gene identifier was assigned to those contigs that showed reliable homology to a sequence in the non-redundant database (E-value cutoff of 1e-20 for significant similarity and 1e-05 for weak similarity). Potential untranslated regions were identified using the program ESTScan \[[@B31]\]. Electronic annotation of contigs -------------------------------- Based on the GO annotation of the closest annotated homolog, contigs were assigned a molecular function, biological process and cellular component from the GO database \[[@B17]\]. To this end, the GenBank annotation files from the GO database were downloaded and parsed for the gene identifier (gi) numbers of previously identified homologs. The cutoff for annotating an A. mexicanumcontig was an E-value of 1e-20. Isolation of the full-length p27^Kip1^gene from the EST sequence ---------------------------------------------------------------- Two EST sequences of the p27^Kip1^gene were sequenced in the EST collection but neither were full-length sequences. To isolate the full-length sequence, 200,000 clones of our arrayed blastema and neural tube libraries were screened by PCR. Briefly, the bacterial library clones in each 384-well plate were pooled, mini-prepped and arrayed into 96-well plates (RZPD, Berlin), resulting in 576 DNA pools. These DNA pools were screened by PCR using the custom SP6 primer (GCACATTAGGCCTATTTAGGTGACA) as a forward primer and a gene-specific p27 reverse primer (TGATTTCCAATGGCTGGTTT). Fifty nanograms of DNA from each pool was used for PCR reactions and PCR cycling was performed at the following conditions: 94°C 2 min, 30 cycles of 94°C 15 sec, 65.5°C 30 sec, 72°C 90 sec, followed by 72°C 7 min). The largest positive band (1.1 kb) was gel purified and sequenced on an ABI377 machine using the SP6 primer. Phylogenetic analysis --------------------- Multiple sequence alignments were done with the program ClustalX \[[@B32]\] using standard parameters. Phylogenetic analysis of mitochondrial 12S rRNA was done using the programs dnadist, phylogenetic analysis of the cyclin B family and the CDK inhibitor family (CKI family) was done using protdist, both from the Phylip package \[[@B33]\]. Trees were calculated with the program fitch from the same software package, using 100 iterations. For the CKI family, only the amino-terminal, CDK-inhibitory domain or the full-length sequences were used for construction of a phylogenetic tree. For the cyclin-B family, only the region overlapping in A. mexicanumcontigs was used for tree construction. Trees were displayed using the program nj-plot \[[@B34]\] for the mitochondrial 12S rRNA tree and unrooted \[[@B34]\] for the CKI- and cyclin B families. Database design --------------- A relational database was created using the open source software MySQL as the database server to store and navigate through resulting sequence contigs and annotations. Scripts connecting the web-based front end to the database were written in the programming language Python. Acknowledgements ================ We thank Wolfgang Zachariae, Ralf Kittler and S Randal Voss for critical reading of the manuscript. We are grateful to Tony Hyman, Albert Poustka and David Drechsel for advice and support. This work was funded by the Max Planck Institute of Molecular Cell Biology and Genetics, and the MeDDrive program of the Medical Faculty, Technical University of Dresden. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Distribution of sequence length. (a)Distribution of read lengths of the sequenced ESTs after quality control. The average read length was 569 bp, corresponding to a peak of between 500 and 600 bp. (b)Distribution of sequence length of assembled contigs. The average length of contigs was 597 bp. (c)Distribution of the number of ESTs per assembled contig. Most of the contigs had one EST. The two largest contigs contained over 400 ESTs (cytochrome coxidase subunit I and 12S rRNA, respectively). ::: ![](gb-2004-5-9-r67-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Homology of A. mexicanumcontigs to protein and nucleotide sequences from other species. (a)Distribution of E-values from the first identified hit in the protein non-redundant database that was used to assign a putative identity to the contig. The majority of contigs identified a protein with an E-value between 1e-20 and 1e-99. In 11% of the cases, the E-value of the first hit was below 1e-100 and can therefore be considered a true ortholog. (b)Distribution of hits in the different sequence databases that were searched sequentially. ::: ![](gb-2004-5-9-r67-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Annotated GO terms and protein domains in the A. mexicanumEST libraries. (a)Gene Ontology electronic annotation in the category \'biological process\' of contigs from A. mexicanum. The largest proportion of annotated contigs was assigned a \'cellular process\' (87%). Of those, five large groups of cellular processes emerged, with \'cell cycle/proliferation\' (13%), \'intracellular signaling\' and \'intracellular transport\' (8% and 15%), \'metabolism\' (17%), \'protein metabolism/modification\' (18%) and \'RNA metabolism\' (13%). (b)Domains associated with cellular processes identified in the A. mexicanumcontig sequence dataset. The largest fraction of contigs was associated with a domain function in \'intracellular transport\', followed by \'RNA-binding and metabolism\' and \'DNA-binding and transcriptional control\'. ::: ![](gb-2004-5-9-r67-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Phylogenetic analysis of the vertebrate cyclin-dependent kinase (CDK) inhibitors (CKIs) p21(Cip1), p27(Kip1) and p57(Kip2). (a)Reference phylogenetic tree of mitochondrial 12S rRNA. The Caudata and Salientia both branch out to build the amphibian group. (b)Unrooted phylogenetic tree of the cyclin B1 gene family. The amphibian cyclin B1 family members form a distinct group. (c)Unrooted phylogenetic tree of the amino-terminal CDK-inhibitory domain of vertebrate p21, p27, p28 and p57, which is conserved between the protein families. p27 of A. mexicanumclearly groups with the p27 proteins from other vertebrates. The amphibian-specific p28-family does not parse with any singe group. Note, however, that unlike the 12S rRNA tree, the A. mexicanumand A. t. tigrinump27 branch out with that of D. rerio. (d)Unrooted, phylogenetic tree of the full-length kinase inhibitor sequences. Using the full-length protein sequences from the CKI families, the p28 family branches off between the p21 and p27 families. (e)Multiple sequence alignment of the amino-terminal, CDK-inhibitory region of the CKI families. The protein sequence of A. mexicanump27 is clearly the ortholog of the p27 family, yet displays higher than expected divergence on the protein level. The same divergence is observed for the ambystomatid p57 proteins. The p28 family has extremely high sequence divergence compared to any other CDKN1 family member. Conserved residues between the three CDKN1 families are highlighted in green and the p28-family in light blue. Residues that differ between ambystomatid sequences and the other vertebrate species are highlighted in the ambystomatid sequences in red. Accession numbers are: NM\_131513 (D. rerioccnb1), NM\_031966 (H. sapiensccnb1), BC041302 (X. laevisccnb1), NM\_172301 (M. musculusccnb1), NM\_171991 (R. norvegicusccnb1), P13351 (X. leavisccnb2), XP\_343420 (R. norvegicusccnb2), P29332 (G. gallusccnb2), NP\_004692 (H. sapiensccnb2), NP\_031656 (M. musculusccnb2), CAC24491 (X. laevisccnb3), P39963 (G. gallusccnb3), CAC94915 (H. sapiensccnb3), NP\_898836 (M. musculusccnb3), AAH56746.1 (D. reriop27A, Drp27A); AAK84219.1 (D. reriop27, Drp27); CN056871.1 (A. t. tigrinump27, Attp27); AAM22491.1 (G. gallusp27, Ggp27); NP\_004055.1 (H. sapiensp27, Hsp27); P46414 (M. musculusp27, Mmp27); NP\_113950.1 (R. norvegicusp27, Rnp27); NP\_000067.1 (H. sapiensp57, Hsp57); P49919 (M. musculusp57, Mmp57); XP\_341967.1 (R. norvegicusp57, Rnp57); CN039016.1 (A. mexicanump57, Amp57); BM489375.1 (G. gallusp57, Ggp57); CK697132.1 (D. reriop57, Drp57); AAH01935.1 (H. sapiensp21, Hsp21); NP\_031695.1 (M. musculusp21, Mmp21); NP\_542960.1 (R. norvegicusp21, Rnp21); AL639561.2 (X. tropicalisp21, Xtp21); BJ065460.1 (X. laevisp21, Xlp21); AAN63876.1 (G. gallusp21, Ggp21); I51683 (X. laevisXic1, XlXic1); BX712320.1 (X. tropicalisp28, Xtp28); TNeu143i03.p1cSP6 (X. tropicalisp28A, Xtp28A); CN033557.1 (A. mexicanump28, Amp28); CN035131.1 (A. mexicanump28A, Amp28A); CN033708.1 (A. mexicanump28B, Amp28B). The scale bar indicates substitutions per site. ::: ![](gb-2004-5-9-r67-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### The Ambystoma mexicanumEST database. A relational database was created as a sequence storage and annotation resource of the sequenced ESTs from A. mexicanum. (a)The main entry site of the EST resource is the contig page, where a subset of the information is available, including the identity of included ESTs, putative identity of the contig, GO annotation including cellular role, biochemical function and cellular component, a list of homologs from different model organisms, and identified conserved domains. Source data are available for all BLAST-based alignments, for external sequence or domain data, and for the complete contig sequence. (b,c)EST information and protein information pages, containing more detailed description of storage information, library source and read length (b). A complete list of homologs and identified conserved domains can be assessed on the protein information page (c). For a more detailed description of the database, see text. ::: ![](gb-2004-5-9-r67-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Some characteristics of the A. mexicanumEST contigs ::: Library Number of sequences Number of contigs (+ singlets) Number of clones in contigs Number of clones in singlets --------------------- --------------------- -------------------------------- ----------------------------- ------------------------------ St18-22 neural tube 7,469 6D tail blastema 9,883 Combined total 17,352 6,377 12,791 4,561 The number of expressed sequence tags sequenced from the two libraries blastema and neural tube, as well as the number of contigs, the number of clones in contigs and the number of clones found in singlets is shown. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Gene definition of the most abundant contigs in the A. mexicanumEST libraries ::: Gene definition Number of clones in contig --------------------------------- ---------------------------- Cytochrome coxidase subunit I 469 12S rRNA 445 Nuclear factor 7 332 Keratin type II 274 Keratin 211 Cytoplasmic β-actin 206 The gene with the highest number of clones identified was cytochrome coxidase subunit I (469 clones in contig), followed by 12S rRNA (445) and nuclear factor 7 (332 clones in contig). ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Contig identities and GenBank identifiers of ESTs unique to A. mexicanum ::: Contig ID GenBank identifier UTR ----------- -------------------- ----- Am\_1065 BI817418.1 Am\_13 BI817561.1 Am\_1868 BI817299.1 Am\_1879 BI817273.1 UTR Am\_1986 BI817397.1 Am\_2156 BI817699.1 UTR Am\_2280 BI817354.1 Am\_242 BI817917.1 Am\_2631 BI817344.1 BI818040.1 BI817371.1 Am\_2695 BI818066.1 UTR Am\_2767 BI817941.1 UTR Am\_2952 BI817736.1 Am\_3070 BI817303.1 Am\_3486 BI817478.1 Am\_3807 BI817992.1 UTR Am\_3828 BI817981.1 BI817250.1 Am\_4598 BI817704.1 Am\_4661 BI817548.1 UTR Am\_4720 BI817653.1 UTR Am\_5031 BI817804.1 UTR Am\_5579 BI818004.1 Am\_5650 BI817315.1 Am\_5742 BI817525.1 UTR Am\_5881 BI818060.1 Am\_6107 BI817553.1 UTR Am\_6128 BI817667.1 UTR Am\_6198 BI817866.1 Am\_646 BI817520.1 BI817607.1 BI817743.1 Am\_6565 BI817313.1 UTR Am\_901 BI817984.1 The table shows contig identities and GenBank identifiers of existing A. mexicanumESTs that do not share any homology to a known protein or nucleotide sequence and can therefore be considered unique. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### The most abundant biological processes assigned to the A. mexicanumcontigs ::: Biological process Total number of contigs \% contigs BL/NT Fisher\'s exact (BL/NT) ------------------------- ------------------------- ------------ --------- ------------------------- Protein metabolism 324 15 116/132 3/1 Metabolism 296 13.7 78/170 0/3 Intracellular transport 268 12.4 59/53 4/5 RNA metabolism 227 10.5 127/45 22/2 Cell cycle 194 9 95/52 5/2 Intracellular signaling 148 6.8 95/65 1/6 DNA metabolism/repair 90 4.1 50/12 3/0 Development 69 3.2 32/27 0/2 Cell-cell communication 81 3.7 24/42 0/6 Differentiation 27 1.5 13/7 2/3 The highest-ranking biological process is \'protein metabolism/modification\' with 15% of contigs assigned. \'Cellular metabolism\', \'intracellular transport\' and \'RNA metabolism\' have all more than 10% of contigs assigned and represent the most abundant gene families in the two libraries. The percentage contigs refers to the number of contigs assigned a biological process. BL: Blastema; NT: Neural tube. ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Common protein domains identified in the A. mexicanumcontigs and comparison to domain occurrences in other vertebrate species ::: Domain A. mexicanum H. sapiens M. musculus X. laevis G. gallus D. rerio --------- ---------------- -------------- --------------- ------------- ------------- ------------ EF-hand 10 319 308 36 48 38 Cyclin 12 60 58 20 9 15 Chromo 5 26 26 8 5 5 Prox1 5 4 2 2 1 2 HLH 8 (1) 167 179 83 70 75 HOX 13 (19) 280 352 196 142 250 PAX 1 (4) 12 31 25 9 13 EGF 10 310 281 26 50 32 SET 2 82 64 4 3 1 RAS 37 220 194 34 11 27 RhoGEF 4 124 98 4 2 3 PH 2 453 374 14 10 18 PX 4 70 74 2 0 3 WD40 39 547 490 63 12 50 Cullin 3 8 20 0 0 2 F-box 2 119 130 11 0 8 HectC 3 64 66 4 2 1 RING 17 374 325 18 16 29 KH 23 (1) 71 52 20 7 10 RRM 101 (2) 443 438 94 23 69 PDZ 8 260 252 17 11 23 Kinase 69 (2) 949 954 210 122 156 LIM 5 128 125 22 19 22 PHD 4 164 122 13 4 3 Numbers in parentheses indicate the number of domains that had been annotated to a protein sequence from A. mexicanumprior to this project. ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Gene families identified that are either involved in cell-cycle control or developmental processes ::: Cellular process Putative ID of contig Contig Expression ------------------ ---------------------------------- ---------- ------------ Cell cycle Cyclin A2 Am\_20 BL unique Cyclin B1 Am\_1031 BL 3x Cyclin B2 Am\_4185 NT unique Cyclin B3 Am\_3173 BL unique Cyclin E1 Am\_38 BL unique Cyclin E2 Am\_91 BL unique Cdk4 Am\_3891 BL unique Polo kinase Am\_1717 BL unique Cdc25A Am\_3678 BL unique p27/Kip1 Am\_4671 NT unique Cdc20 Am\_2213 BL unique Cdh1 Am\_1148 BL unique Development Wnt8 Am\_384 BL unique Wnt5 Am\_642 BL unique FGFR4a Am\_1393 BL unique Sonic hedgehog Am\_3741 BL unique Activin receptor type II Am\_3590 BL unique TGF-β Am\_4990 NT unique BMP-1 Am\_4639 NT unique Cdx1 Am\_875 BL unique Cdx2 Am\_387 BL unique HoxA2 Am\_2387 BL unique HoxB13 Am\_4865 NT unique HoxC4 Am\_3998 BL unique HoxC8 Am\_2910 BL unique Pax6 Am\_2945 BL unique Smad5 Am\_1420 BL unique Smad8 Am\_4665 NT unique Retinoblastoma binding protein 2 Am\_2723 BL unique Beta-catenin Am\_699 BL 3x Zic5 Am\_2068 BL unique Frizzled 2 Am\_3243 BL unique Frizzled 5 Am\_3451 BL unique Frizzled 7 Am\_2334 BL unique TRIP12 Am\_6416 NT unique The identifier of the A. mexicanumcontig is given in the third column. The expression pattern as determined by in silicodifferential display is shown in column 4. ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Occurrence of CKI-family members in different vertebrate species ::: Human Zebrafish Fugu Xenopus tropicalis Ambystoma mexicanum -------------------- ------- ----------- ------ ---------------------- ----------------------- CDKN1A (p21) \+ -\* \+ \+ -\* CDKN1B (p27^Kip1^) \+ \+ \+ -^†^ \+ CDKN1C (p57) \+ \+ \+ -^†^ \+ p28^Kix1^ \- \- \- +^‡^ +^‡^ \* Genes most likely present, yet not identified due to limited sequence information; † genes not present in genomic sequence information; ‡ genes so far only present in amphibian species. Databases searched were the human, mouse, rat, fugu, zebrafish and X. tropicalisgenome databases, and the EST databases for X. laevis, X. tropicalis, zebrafish, A. mexicanumand A. tigrinum. :::	PubMed Central	2024-06-05T03:55:48.513728	2004-8-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522874/", "journal": "Genome Biol. 2004 Aug 13; 5(9):R67", "authors": [ { "first": "Bianca", "last": "Habermann" }, { "first": "Anne-Gaelle", "last": "Bebin" }, { "first": "Stephan", "last": "Herklotz" }, { "first": "Michael", "last": "Volkmer" }, { "first": "Kay", "last": "Eckelt" }, { "first": "Kerstin", "last": "Pehlke" }, { "first": "Hans Henning", "last": "Epperlein" }, { "first": "Hans Konrad", "last": "Schackert" }, { "first": "Glenis", "last": "Wiebe" }, { "first": "Elly M", "last": "Tanaka" } ] }
PMC522875	Background ========== MicroRNAs constitute a large class of small regulatory RNAs \[[@B1]\]. Their mechanism of action and the scope of their biological roles are beginning to be understood. The first two microRNAs were discovered as the products of heterochronic genes that control developmental timing in Caenorhabditis elegans\[[@B2]\]. In heterochronic mutants, the timing of specific developmental events in several tissues is altered relative to the timing of events in other tissues. These defects result from temporal transformations in the fates of specific cells; that is, certain cells acquire fates normally expressed by cells at other developmental stages. The molecular characterization of the heterochronic gene lin-4led to the surprising discovery that this gene encodes a 21-nucleotide non-coding RNA that regulates the translation of lin-14mRNA through base-pairing with the lin-143\' UTR \[[@B3],[@B4]\]. A second heterochronic gene, let-7, encodes another small non-coding RNA that is conserved in flies and mammals \[[@B5]\]. Biochemical and bioinformatic approaches have identified many genes that encode microRNAs in C. elegans, plants, Drosophila melanogasterand mammals \[[@B6]-[@B18]\]. Like the lin-4and let-7genes, other microRNAs encode 21-25-nucleotide RNAs derived from longer transcripts that are predicted to form stem-loop structures. More than 200 microRNAs are encoded by the human genome \[[@B8],[@B14]\]. The biological roles of microRNAs are poorly understood. In C. elegans, lin-4and let-7act in developmental timing, and the microRNA lsy-6controls neuronal asymmetry \[[@B19]\]. In Drosophila, the microRNAs bantam and mir-14 act in the regulation of cell growth and cell death \[[@B20],[@B21]\]. The mouse microRNA miR-181 has been implicated in the modulation of hematopoietic differentiation, and other mammalian microRNAs have been suggested to play roles in cancer \[[@B22],[@B23]\]. Mature microRNAs are excised from a stem-loop precursor that itself can be transcribed as part of a longer primary RNA (pri-miRNA) \[[@B24]\]. The pri-miRNA appears to be processed by the RNAse Drosha in the nucleus, cleaving the RNA at the base of the stem-loop \[[@B25]\]. This cut defines one end of the microRNA. The precursor microRNA is then exported by Ran-GTP and Exportin-5 to the cytoplasm, where it is further processed by the RNAse Dicer \[[@B26],[@B27]\]. Dicer recognizes the stem portion of the microRNA and cleaves both strands about 22 nucleotides from the base of the stem \[[@B25]\]. The two strands in the resulting double-stranded (ds) RNA are differentially stable, and the mature microRNA resides on the strand that is more stable \[[@B28],[@B29]\]. Mature microRNAs can be found associated with the proteins eIF2C2 (an Argonaute-like protein), Gemin2 and Gemin3 and are thought to act in a protein-RNA complex with these and maybe other proteins \[[@B17],[@B30]\]. The animal microRNAs studied so far act by reducing the levels of proteins from genes that encode mRNAs with sites partially complementary to microRNAs in their 3\' UTRs \[[@B4],[@B31]\]. The mechanism responsible is not understood in detail \[[@B32]\]. In contrast, although some plant microRNAs with partially complementary target sites also act by preventing translation, the majority studied so far cause the cleavage of target mRNAs at sites perfectly complementary to the microRNAs \[[@B33]-[@B38]\]. Determining spatial and temporal patterns of microRNA expression should yield insight into the biological functions of microRNAs. As the number of microRNAs identified has increased rapidly, the need for a method that allows for the parallel detection of microRNA expression has become apparent. Recent studies used a dot-blot technique to study 44 mouse microRNAs and northern blotting analysis to study 119 microRNAs from mouse and human organs \[[@B39],[@B40]\]. In this study we cloned microRNAs from rat and monkey brains, developed a microRNA labeling method and used a microarray to monitor expression of microRNAs during mouse brain development. We determined the temporal expression pattern of 138 microRNAs in the mouse brain and found that the levels of 66 microRNAs changed significantly during development. We identified sets of genes with similar expression patterns, including genes that peaked in expression at different stages of development. More generally, the microRNA microarray we have developed can be used to determine the expression of all known microRNAs simultaneously under any set of experimental conditions or constraints. Results and discussion ====================== Identification of microRNAs from developing rat and monkey brains ----------------------------------------------------------------- To analyze microRNAs expressed in the developing mammalian brain, we cloned small 18-26-nucleotide RNAs from the neocortex and hippocampus of a 12-day postnatal rat (Rattus norvegicus) and from the cerebral wall of a 114-day-old fetal rhesus monkey (Macaca mulatta) (Table [1](#T1){ref-type="table"}). In both species, by these stages most neurons have been generated and have begun synaptogenesis \[[@B41],[@B42]\]. We identified a total of 1,451 sequences, 413 of which correspond to microRNA sequences on the basis of their potential to generate stem-loop precursors as predicted from corresponding sequences in the rat and/or human genomes. In all cases but one, the microRNAs we identified corresponded to known microRNAs from other species and defined 68 unique microRNAs (Table [1](#T1){ref-type="table"} and Additional data file 1). One of these microRNAs is new: it differs in sequence from any microRNA previously described and is conserved in the mouse and human genomes. We named this new microRNA rno-miR-421 (Figure [1](#F1){ref-type="fig"} and Additional data file 2). As observed in similar studies, in addition to microRNAs a number of candidate small RNAs that do not fulfill all criteria of a microRNA were cloned (Additional data file 3) \[[@B9],[@B43]\]. Of the 52 rat microRNA sequences we cloned, 27 had previously been cloned from rat primary cortical neurons \[[@B11]\]. For 21 of the 52 microRNAs from rat and 14 of the 40 microRNAs from monkey we isolated only a single clone, indicating that our surveys are not saturated. By contrast, we isolated microRNA miR-124a 19 times from rat and 97 times from monkey. Mouse miR-124a as well as miR-128, miR-101 and miR-132 have been reported to be expressed specifically in brain \[[@B15]\]. We found that rat miR-138 also was expressed only in brain (Additional data file 4). MicroRNA microarrays for the study of temporal and spatial patterns of microRNA expression ------------------------------------------------------------------------------------------ Previous analyses of microRNA expression have relied on dot blots, northern blots and cloning strategies \[[@B8],[@B11]-[@B14],[@B18],[@B39],[@B40]\]. A highly scalable approach using a microarray would facilitate the analysis of microRNA expression patterns for a large number of samples and is feasible now that many mammalian microRNAs have been identified. We arrayed 138 oligonucleotides complementary to microRNAs (probes) corresponding to the 68 mammalian microRNAs we isolated from rat and monkey brains, to 70 mammalian microRNAs isolated by others from a variety of mouse tissues and mammalian cell lines, and to predicted microRNAs. In addition, we included a set of control probes as well as 19 probes corresponding to presumptive small RNAs that we and others identified but that do not satisfy all the criteria for a microRNA (see below and Additional data file 5). Probes had a free amine group at the 5\' terminus and were printed onto amine-binding glass slides and covalently linked to the glass surface. All probes were printed in quadruplicate (Additional data file 5). We developed a method for preparing microRNA samples for microarray analysis. Several methods for mRNA sample labeling for microarray analysis have been described \[[@B44]-[@B47]\], but none is suitable for labeling RNAs as small as microRNAs. To fluorescently label small RNAs we adapted strategies for RNA ligation and reverse transcription PCR (RT-PCR) devised for microRNA cloning \[[@B12]-[@B14]\]. Briefly, 18-26-nucleotide RNAs were size-selected from total RNA using denaturing polyacrylamide gel electrophoresis (PAGE), oligonucleotide linkers were attached to the 5\' and 3\' ends of the small RNAs and the resulting ligation products were used as templates for an RT-PCR reaction with 10 cycles of amplification. The sense-strand PCR primer had a Cy3 fluorophore attached to its 5\' end, thereby fluorescently labeling the sense strand of the PCR product. The PCR product was denatured and then hybridized to the microarray. As in microarray analysis, the labeled sample used for hybridization is referred to as the target. Significant biases in amplification, a problem when amplifying heterogeneously sized mRNAs, are less likely in the case of microRNAs because of their short uniform lengths. MicroRNA cloning frequencies obtained using a similar amplification strategy correlate well with expression levels as assayed by quantitative northern blots \[[@B7]\]. Because RNA is amplified before hybridization, relatively low amounts of starting material may be used with this method \[[@B8],[@B11]-[@B14],[@B18],[@B39],[@B40]\]. We optimized the conditions for hybridization to our microarray. The small sizes of microRNAs leave little opportunity for oligonucleotide (array probe) design to achieve homogeneous probe-target melting temperatures. Differences in melting temperatures are expected to result in greater nonspecific binding if hybridizations are performed at low temperatures (to allow the detection of probe-target pairs with low melting temperatures) and in less specific binding if hybridizations are performed at high temperatures (to specifically detect probe-target pairs with high melting temperatures). To assess this issue we included control probes with two internal mismatches on the microarray for a subset of the microRNA probes (Additional data file 5). We tested a range of hybridization temperatures, and, on the basis of the signal of microRNA probes versus control probes, we determined that a hybridization temperature of 50°C was a reasonable compromise between sensitivity and specificity (data not shown). Even at 50°C, specificity as assayed by comparing microarray spot signal intensities from matched and mismatched probes varied among the microRNAs assayed. As expected, specificity at 50°C was negatively correlated with calculated melting temperatures (Figure [2a](#F2){ref-type="fig"}). In all cases the cumulative signal from 10 hybridizations for the mismatched probe was equal to or lower than that for the microRNA probe, but differences in the ratio of the matched to mismatched probe signal ranged widely (Figure [2a](#F2){ref-type="fig"}). Given these data, we do not expect the microRNA microarray to distinguish reliably between microRNAs that have only one or a few mismatches. This limitation is alleviated somewhat by the fact that for most microRNAs that have been identified the most closely related paralogs differ by five mismatches or more (Figure [2b](#F2){ref-type="fig"}). The signal from a mismatched control probe is likely to be caused by cross-hybridization with the microRNA for which it was designed, as other control probes corresponding to unrelated mRNA subsequences or synthetic probes that do not correspond to known microRNAs did not show signals above background (Additional data file 5). Microarray results for closely related microRNAs should be interpreted with caution, as differences in the apparent expression of a given microRNA could be dampened or exaggerated depending on the expression of the paralogs (Figure [2a](#F2){ref-type="fig"}). To determine the detection range of the microarray, we synthesized three artificial RNAs with the characteristics of microRNAs. These RNAs were phosphorylated RNA oligonucleotides of 20-23 bases; their sequences were chosen at random and were without any significant sequence similarity to known mammalian microRNAs (see Additional data file 5 for details). We titrated these RNAs into total mouse RNA samples, labeled them and hybridized them to a microarray that in addition to microRNA probes included probes corresponding to these three RNAs, called syn1, syn2 and syn3. Figure [2c](#F2){ref-type="fig"} shows the correlation between the amount of the RNAs and the microarray signal intensities. For comparison, the background signal for the array is also shown. All three RNAs were reliably detected at levels as low as 0.1 fmoles. The dynamic range of the array was from 0.1 fmoles to at least 10 fmoles, or two orders of magnitude. Analysis of microRNA expression during mouse brain development -------------------------------------------------------------- We isolated small RNAs from mice at five developmental stages: embryonic days 12.5 and 17.5 (E12.5 and E17.5), postnatal days 4 and 18 (P4 and P18) and 4-month-old adults. E12.5-E17.5 spans a period of major neuronal proliferation and migration in the mouse brain, in particular the birth and subsequent migration of most neurons in the ventricular zone epithelium \[[@B48]\]. Between postnatal days P4 and P18, major sensory inputs are established. For example, eye opening occurs around P13 and is thought to result in activity-dependent neuronal remodeling \[[@B49]\]. We purified and size-selected RNA from whole mouse brains. For each sample, the products of four independent RNA amplifications based on two independent RNA ligations were hybridized to the array. A detailed description of our analysis of the microarray data is presented in Additional data file 5. Of the 138 microRNAs and 19 small RNAs represented by the probe set, 116 (74%) were expressed robustly (more than 75-fold over the level of background controls) at least at one time point. Of these, 83 (71%) changed significantly during the period surveyed (analysis of variance, ANOVA, p\< 0.001) and 66 (57%) changed more than twofold. Of the microRNAs we cloned from rat and monkey and for which probes against the corresponding mouse homologs were present on the microarray, we detected 97% robustly. We grouped microRNAs that changed more than twofold in expression during the period analyzed using a hierarchical clustering algorithm (Figures [3a](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}) \[[@B50]\]. A group of microRNAs peaked at each of the developmental time points. The signal from 34 of the 66 probes that changed more than twofold peaked in the fetus (E12.5 and E17.5), suggesting roles in early development (Figure [4a](#F4){ref-type="fig"}). Nine and eleven microRNAs peaked during the neonate (P4) and juvenile (P18) stages, respectively. Twelve microRNAs had the highest signals at the adult stage (Figure [4b](#F4){ref-type="fig"}). These data indicate that murine brain development involves a wave of expression of sequential classes of microRNAs (Figure [3a](#F3){ref-type="fig"}). We also grouped the developmental time points according to their microRNA expression pattern using hierarchical clustering. We found that samples from stages that are developmentally proximal had the most similar microRNA expression patterns (Figure [3b](#F3){ref-type="fig"}), indicating that a microRNA expression profile can be a marker of developmental stage. Examination of the temporal clusters revealed that probes with similar sequences showed correlated expression, as exemplified by miR-181a, miR-181b, miR-181c, smallRNA-12 (Figure [4a](#F4){ref-type="fig"}) and miR-29a, miR-29b and miR-29c (Figure [4b](#F4){ref-type="fig"}), respectively. Given our observation that the microRNA microarray can detect mismatched sequences, it is possible that this correlation among closely related family members is an artifact of hybridization. We found that four of the 66 RNAs that changed more than twofold were small RNAs rather than microRNAs. The temporal regulation of these small RNAs indicates that they may play a role during development. Several mouse microRNAs are clustered closely in the genome, suggesting that they might be expressed from a single precursor transcript or at least share promoter/enhancer elements. We searched all known microRNA clusters in the mouse genome to attempt to identify coordinately controlled clustered microRNAs. We sought clusters with the following features: first, the clustered microRNAs are not all members of the same family; second, the microRNAs have no or few paralogs; and third, the microRNAs are detected robustly on our microarray and their expression changes significantly during the timecourse studied. The mir-17cluster on chromosome 14 fulfills all these criteria. Figure [4c](#F4){ref-type="fig"} shows that the expression of all six microRNAs in this cluster is indeed highly co-regulated. Validation of microarray results using northern blots ----------------------------------------------------- To validate our microarry results, we performed northern blots of eight microRNAs that were robustly expressed at least at one point during development according to our microarray data. The relative changes of microRNA expression assayed using microarray analysis and northern blots were consistent (Figure [5](#F5){ref-type="fig"}). For example, on a northern blot miR-29b was almost undetectable at the embryonic and P4 stages but appeared at P18 and was strongly expressed in the adult. The microarray data showed a similar pattern. In only a few cases did there seem to be discrepancies; for example, relative levels of expression of miR-138 at P4 compared to adult differed between the northern blots and the microarrays. As is the case for mRNAs, small differences may be seen between the methods and northern blot analysis is superior to microarrays for quantitative analysis \[[@B51]\]. Nonetheless, microarrays offer a high-throughput method that generally captures changes in microRNA expression. Conclusions =========== Here we describe the development of a microarray technology for profiling the expression of microRNAs and other small RNAs and apply this technology to the developing mammalian brain. Recently, Krichevsky et al.described the temporal expression of 44 microRNAs during mouse brain development \[[@B39]\]. Their study used a dot-blot array approach and direct labeling of microRNAs using radioactivity instead of a glass microarray and RT-PCR/fluorescent labeling, as we used in our study. Despite differences in sample selection as well as in the number of microRNAs analyzed, there is good agreement between the overlapping aspects of the two datasets. Our strategy has the potential to be highly scalable, allowing high-throughput analysis of samples with limiting starting material. MicroRNA microarrays offer a new tool that should facilitate studies of the biological roles of microRNAs. We speculate that some of the developmentally regulated microRNAs we describe in this report play roles in the control of mammalian brain development, possibly by controlling developmental timing, by analogy to the roles of the lin-4and let-7microRNAs in C. elegans. Materials and methods ===================== MicroRNA cloning ---------------- We isolated RNAs and cloned microRNAs from R. norvegicusand M. mulattausing methods described previously \[[@B13]\], except that the samples were not dephosphorylated during the cloning procedure. Microarray printing and hybridization ------------------------------------- Microarray probes were oligonucleotides (named EAM followed by a number) with sequences complementary to microRNAs. Each probe was modified with a free amino group linked to its 5\' terminus through a 6-carbon spacer (IDT) and was printed onto amine-binding slides (CodeLink, Amersham Biosciences). Control probes contained two internal mismatches resulting in either C-to-G or T-to-A changes (Additional data file 6). Printing and hybridization were done using the protocols from the slide manufacturer with the following modifications: the oligonucleotide concentration for printing was 20 μM in 150 mM sodium phosphate pH 8.5, and hybridization was at 50°C for 6 h. Printing was done using a MicroGrid TAS II arrayer (BioRobotics) at 50% humidity. Sample and probe preparation ---------------------------- Whole brains from three to eight C57BL/6 mice were pooled. Starting with 250 μg of total RNA for each time point, 18-26-nucleotide RNA was purified on denaturing PAGE gels. The samples were divided, and the following cloning steps were done independently twice for each time point. 3\' and 5\' adaptor oligonucleotides were ligated to 18-26-nucleotide RNA followed by reverse transcription, essentially as described for microRNA cloning \[[@B13]\]. Briefly, a RNA-DNA hybrid 5\'-pUUUaaccgcgaattccagt-idT-3\' (Dharmacon: X, RNA; x, DNA; p, phosphate; idT, inverted \[3\'-3\' bond\] deoxythymidine) was ligated to the 3\' end and 5\'-acggaattcctcactAAA-3\' (Dharmacon) was ligated to the 5\' end. The ligation products were divided into two aliquots, and the following steps were done independently twice for each time point. Ligation products were reverse transcribed and amplified by 10 rounds of PCR (40 sec at 94°C, 30 sec at 50°C, 30 sec at 72°C). For PCR, the oligonucleotides used were: oligo1 5\'-Cy3-ACGGAATTCCTCACTAAA-3\' and oligo2 5\'-TACTGGAATTCGCGGTTAA-3\'. The PCR product was precipitated, washed and resuspended in hybridization buffer (5× SSC, 0.1% SDS, 0.1 mg/ml sheared denatured salmon sperm DNA). Data acquisition and analysis ----------------------------- Microarray slides were scanned using an arrayWoRx biochip reader (Applied Precision), and primary data were analyzed using the Digital Genome System suite (MolecularWare) and Spotfire DecisionSite (Spotfire). Cluster analysis was performed using the CLUSTER/TreeView software \[[@B50]\]. For details concerning microarray data analysis see Additional data file 5. The predicted stem-loop RNA structures were generated using the mfold (version 3.1) software \[[@B52]\]. Northern blots -------------- Northern blots were performed as described \[[@B14]\]. Twenty micrograms of total RNA were loaded per lane. A probe for the mouse U6 snRNA (5\'-TGTGCTGCCGAAGCGAGCAC-3\') was used as loading control. The probes for the northern blots had the same sequences as the corresponding EAM oligonucleotides printed on the microarray (see Additional data file 6). The blots were stripped by boiling for 5 min in distilled water and reprobed up to four times. The probes used were: EAM119 (miR-29b), EAM125 (miR-138), EAM224 (miR-17-5p), EAM234 (miR-199a), EAM131 (miR-92), EAM109 (miR-7), EAM150 (miR-9) and EAM103 (miR-124a). Additional data files ===================== The following additional data files are available with the online version of this article. A file (Additional data file [1](#s1){ref-type="supplementary-material"}) with details of rat microRNA precursors: using the assembly of the rat genome \[[@B55]\] we identified candidate genomic locations for all of our rat microRNAs that have orthologs in the mouse but that have not been described previously for the rat. An alignment of the top BLAST hit of each mouse microRNA precursor sequence (The miRNA Registry, Release 3.2) against the rat genome sequence (public release draft genome assembly, version 3.1) is shown in this file. In addition, predicted precursor secondary structures are shown for each rat microRNA gene for which the precursor sequence differs from that of the corresponding mouse precursor. We used the mfold algorithm to make secondary structure predictions \[[@B52]\]. A file (Additional data file [2](#s2){ref-type="supplementary-material"}) containing precursor sequences and secondary structure predictions for the novel microRNA miR-421. An alignment of the predicted precursor sequences from human and mouse of the novel microRNA miR-421, which we identified from rat, is shown. The cloned sequence (corresponding to the mature microRNA) is shown in bold. The single mismatch is indicated by an asterisk (\). The accession number for each sequence is given. The predicted secondary structures for the corresponding genes from mouse and human are also shown. The human and mouse genomic sequences for candidate miR-421 precursors are identical. The sequence of the mature microRNA is colored in red. The residue in the rat genomic sequence that is different from the mouse and human genomic sequences is indicated. We used the mfold algorithm to make secondary structure predictions \[[@B52]\]. We have not been able to detect miR-421 in mouse brain using northern blots. A file (Additional data file [3](#s3){ref-type="supplementary-material"}) with details of other small RNAs cloned from rat and monkey brains. We cloned 13 small RNAs that do not satisfy all criteria to be considered microRNAs. One, small RNA-1 from monkey, is present in the mouse genome and has a predicted stem-loop precursor sequence characteristic of microRNAs. However, the predicted stem-loop ends on the final base of the microRNA, which is not typical for microRNAs. To be conservative, we refer to this RNA as a small non-coding RNA. A northern blot for smallRNA-1 revealed a high molecular weight band that may represent a precursor RNA (data not shown). Since there is no perfect match to smallRNA-1 in the current release of the human sequence, a presumptive precursor based on mouse genomic sequence is shown. The cloned sequence is highlighted in red. The other small RNAs are not predicted to form stem-loop structures. A file (Additional data file [4](#s4){ref-type="supplementary-material"}) showing rno-miR-138 brain specific expression. A northern blot shows that rno-miR-138 expression was restricted to brain. The probe was identical to EAM125. Total RNA isolated from various adult rat tissues (Ambion) was size-separated on denaturing PAGE (12 μg per lane), transferred to a nylon membrane and used for hybridization. Equal loading was verified using a probe for U6 snRNA (data not shown). A file (Additional data file [5](#s5){ref-type="supplementary-material"}) with details of microarray design and data analysis. And a file (Additional data file [6](#s6){ref-type="supplementary-material"}) with a summary of the microRNA microarray data. Oligonucleotide sequences correspond to probes on the array. MicroRNA names were obtained from the miRNA registry (Release 3.2) or if not available these were obtained from NCBI. Probes named smallRNA-1 through -13 correspond to unique small RNAs that we cloned but that did not correspond to known microRNAs and did not have perfect matches in the current release of the rat genome sequence. Column A indicates whether the probe is complementary to a microRNA or to one of the small RNAs we cloned (\"-\" = no, \"+\" = yes). Column B indicates whether the probe is complementary to a mouse microRNA. Oligonucleotides with a \"-\" in column A were either controls or sequences that were submitted to public databases as microRNAs and later found not to encode microRNAs. In a few cases we printed probes that represented the same microRNA twice, but we analyzed the data from only one of these probes. The probes we did not analyze had no labels in columns A, B, and C. Melting temperatures were calculated using the nearest neighbors method \[[@B53]\]. Data for the five time points of mouse brain development (E12.5, E17.5, P4, P18 and adult) are shown. Microarray data were derived as described (Additional data file [5](#s5){ref-type="supplementary-material"}). Briefly, data correspond to mean spot intensities averaged over quadruplicates (on each array) and four independent hybridizations. SEM refers to the standard error of the mean. microRNAs labeled with ^%^are not in the current release of the miRNA registry but are deposited in NCBI and described elsewhere \[[@B8],[@B15],[@B18]\]. An Excel file (Additional data file [7](#s7){ref-type="supplementary-material"}) with the primary microarray data corresponding to the E12.5 time point; an Excel file (Additional data file [8](#s8){ref-type="supplementary-material"}) with the primary microarray data corresponding to the E17.5 time point; an Excel file (Additional data file [9](#s9){ref-type="supplementary-material"}) with the primary microarray data corresponding to the P4 time point; an Excel file (Additional data file [10](#s10){ref-type="supplementary-material"}) with the primary microarray data corresponding to the P18 time point; an Excel file (Additional data file [11](#s11){ref-type="supplementary-material"}) with the primary microarray data corresponding to the Adult time point. All data in Additional files 7-11 were exported from the Digital Genome System suite (MolecularWare). The primary microarray data will also be submitted to the Gene Expression Omnibus (GEO) database \[[@B56]\]. Supplementary Material ====================== ::: {.caption} ###### Additional data file 1 A file with details of rat microRNA precursors ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 2 A file containing precursor sequences and secondary structure predictions for the novel microRNA miR-421 ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 3 A file with details of other small RNAs cloned from rat and monkey brains ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 4 A file showing rno-miR-138 brain specific expression ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 5 A file with details of microarray design and data analysis ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 6 A file with a summary of the microRNA microarray data ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 7 An Excel file with the primary microarray data corresponding to the E12.5 time point ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 8 An Excel file with the primary microarray data corresponding to the E17.5 time point ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 9 an Excel file with the primary microarray data corresponding to the P4 time point ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 10 An Excel file with the primary microarray data corresponding to the P18 time point ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 11 An Excel file with the primary microarray data corresponding to the Adult time point ::: ::: {.caption} ###### Click here for additional data file ::: Acknowledgements ================ We thank Nelson Lau for help with microRNA cloning and northern blots. We also thank Sanchita Bhattacharya, Xuan Shirley Li and Sean Milton from the MIT BioMicroCenter for help with microarray printing and data analysis and Lucila Scimone and Justin Lamb for help with the manuscript and discussions. We thank David Bartel for critical reading of the manuscript. E.A.M. was supported by an EMBO fellowship (ALTF-153-2000) and a Wellcome Trust International Prize Travelling Research Fellowship (GR061641). This work was supported by the Howard Hughes Medical Institute (H.R.H.) and by grants from the National Institute of Neurological Disorders and Stroke (P.R.) and grant R01EY06039 from the National Eye Institute (M.C.P.). H.R.H is an Investigator of the Howard Hughes Medical Institute. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Predicted stem-loop structure of a novel mammalian microRNA, rno-miR-421. The stem-loop structure was predicted from sequences adjacent to rno-miR-421 in the rat genome. The cloned (mature) sequence is shown in red. The predicted secondary structure and the free energy calculation (ΔG, kcal/mole) were generated by the mfold software \[52\]. ::: ![](gb-2004-5-9-r68-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### MicroRNA microarray specificity and quantification. (a)Specificity was assayed using a set of 23 microRNA and mismatched probe pairs (two mismatches). Average mean spot intensities from 10 independent hybridizations at 50°C were added to give a total signal for probes corresponding to a given microRNA as well as for probes with two mismatches to the microRNA. Mismatch probe design and sequences are described in Additional data file 6. A specificity index was calculated as 100 × (probe signal - mismatched probe signal)/probe signal. Melting temperatures for the microRNA probes were calculated using the nearest neighbors method \[54\]. The specificity index is plotted against the calculated melting temperature for each microRNA probe pair. Correlation of melting temperature and specificity index is significant (p= 0.004, Student\'s t-test). (b)Number of mismatches between microRNAs based on all known mouse microRNAs (the miRNA Registry 3.0 \[53\]). Each microRNA was aligned pairwise to every other microRNA and was assigned to the group (number of mismatches) corresponding to the least number of mismatches to another microRNA. (c)Quantification of microarray data using three synthetic RNAs: syn1, syn2 and syn3. Each data point is the average of two independent labelling/hybridization reactions. Probes for the three synthetic RNAs were printed in quintuplicate on the microarray. RNAs were used at 0.025, 0.1, 0.375, 0.75, 2.5, 5 and 10 fmoles. For comparison, the background signal of the array is shown. For more details, see Additional data file 5. ::: ![](gb-2004-5-9-r68-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Profile of microRNA expression in the developing mouse brain. (a)Relative expression levels for the 66 microRNAs that changed significantly (ANOVA, p\< 0.001) and more than twofold are shown in five columns corresponding to the five time points. Colors indicate relative signal intensities. The microRNA expression profile was sorted using a hierarchical clustering method, and major clusters are shown ordered according to the time that expression peaks. Gene names and a quantitative description of microRNA expression levels are presented in Additional data file 6. (b)Developmental time points were grouped using the same hierarchical clustering method and gene set as in (a). ::: ![](gb-2004-5-9-r68-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Examples of co-regulated microRNAs. (a)MicroRNAs with a sharp peak at the E12.5 stage. Methods were as described for Figure 3. Brackets indicate closely related sequences. (b)MicroRNAs with a single sharp peak at the adult stage. (c)Co-regulation of microRNAs derived from the mir-17cluster from chromosome 14. To compare signal intensities, data were transformed to give a mean of 0 and a standard deviation of 1. ::: ![](gb-2004-5-9-r68-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Comparison of microarray data with representative developmental northern blots of microRNAs. Northern blots were prepared and microarray analysis was done using the same starting material. For each microRNA, northern blots (left panel) and the microarray hybridization signals (right panel) are shown. Quantification of northern blots is also shown (middle panel). Y-axis for the microarray data refers to the averaged mean signal intensities (× 10^-3^), and error bars are standard errors of the mean. Northern blots were done using 20 μg of total RNA in each lane. Because northern blots were exposed for different lengths of time, the intensities of the signals on northern blots cannot be directly compared to those from the microarrays. A probe against U6 snRNA was hybridized to the same blots for comparison. ::: ![](gb-2004-5-9-r68-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Identity, frequency and size range of microRNAs cloned from the cortex and hippocampus of 12-day postnatal R. norvegicusand the cortex of a 114-day old M. mulattafetus ::: Rattus norvegicusmicroRNAs Macaca mulattamicroRNAs ------------------------------ --------------------------- ------- --------------------- ----- ------- rno-miR-421 2 21 rno-let-7a 3 22 mml-let-7a 15 21 mml-let-7a or c 1 18 rno-let-7b 1 23 mml-let-7b 20 22-23 rno-let-7c 10 22 mml-let-7c 9 21-22 rno-let-7d 1 22 mml-let-7d 1 22 mml-let-7e 3 20-22 mml-let-7f 3 22 mml-let-7g 2 22 rno-let-7i 1 22 mml-let-7i 2 22 rno-miR-7 5 21 mml-miR-7-1 1 22 rno-miR-9 2 23 mml-miR-9 9 21-23 rno-miR-16 2 22 mml-miR-16 2 22 rno-miR-17-5p 3 23 mml-miR-17-5p 2 22-23 rno-miR-24 6 21-22 mml-miR-26a 3 21-22 rno-miR-26b 1 22 rno-miR-28 1 22 rno-miR-29a 4 22 rno-miR-29b 7 22-23 rno-miR-29c 2 20,22 rno-miR-30b 1 22 mml-miR-30b 2 22 rno-miR-30c 3 23-24 mml-miR-30c 1 21 mml-miR-33 2 20 rno-miR-92 2 22 mml-miR-92 rno-miR-93 1 23 rno-miR-99a 1 21 mml-miR-99a 4 20-22 rno-miR-99b 2 21,22 mml-miR-99b 2 22 mml-mir-100 1 22 rno-miR-101b 1 22 rno-miR-103 3 23 mml-miR-103 2 22-23 mml-miR-103 or 107 1 21 rno-miR-124a 19 19-22 mml-miR-124a 97 18-23 rno-miR-125a 2 22,24 mml-miR-125a 4 22-23 rno-miR-125b 12 21-22 mml-miR-125b 17 20-22 mml-miR-126 1 21 mml-miR-126\ 1 22 rno-miR-127 1 20 rno-miR-128a 3 21-22 mml-miR-128a 9 22 rno-miR-128a or b 2 21 mml-miR-128a or b 17 18-21 rno-miR-128b 1 21 mml-miR-128b 8 22 rno-miR-129 2 21-22 mml-miR-129-2 1 22 rno-miR-130a 1 22 rno-miR-132 6 22 rno-miR-136 2 23 mml-miR-136 1 23 mml-miR-137 1 23 rno-miR-138 5 23-24 rno-miR-139 1 23 mml-miR-140 1 22 rno-miR-140\* 1 22 rno-miR-142-3p 1 23 rno-miR-145 1 23 mml-miR-145 2 22 rno-miR-146 2 23 mml-miR-149 2 23 rno-miR-150 4 22-23 rno-miR-154 1 22 mml-miR-181a or 213 4 20-25 mml-miR-181b 1 24 mml-miR-181c 1 21 rno-miR-185 2 22-23 mml-miR-185 1 23 rno-miR-191 3 23-24 mml-miR-195 1 22 rno-miR-213 1 22 mml-miR-221 3 22-23 rno-miR-300 1 21 rno-miR-323 1 22 rno-miR-324 4 23 rno-miR-325 1 22 rno-miR-338 5 23 rno-miR-342 1 25 rno-miR-345 1 22 Total 152 Total 261 The rat (rno) and monkey (mml) microRNA names are indicated. Two microRNA names are assigned to the same clone when the cloned sequence is too short to distinguish between the microRNAs. mml-miR-7 and mml-miR-129 are encoded by three and two distinct genomic loci, respectively, although the sequences immediately adjacent to these microRNA sequences differ. The sequences we cloned for mml-miR-7-1 and mml-miR-129-2 were one base longer than that shared by the microRNAs, allowing us to determine the loci from where they originated, as indicated by -1 and -2. Notation follows the miRNA registry guidelines \[53\]. :::	PubMed Central	2024-06-05T03:55:48.519784	2004-8-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522875/", "journal": "Genome Biol. 2004 Aug 31; 5(9):R68", "authors": [ { "first": "Eric A", "last": "Miska" }, { "first": "Ezequiel", "last": "Alvarez-Saavedra" }, { "first": "Matthew", "last": "Townsend" }, { "first": "Akira", "last": "Yoshii" }, { "first": "Nenad", "last": "Šestan" }, { "first": "Pasko", "last": "Rakic" }, { "first": "Martha", "last": "Constantine-Paton" }, { "first": "H Robert", "last": "Horvitz" } ] }
PMC522876	Background ========== Microarrays allow the interrogation of the transcriptome, the set of genes transcribed in a particular cell type under a particular condition \[[@B1]\]. Arrays are particularly potent tools when their coverage is relatively comprehensive, based on a completed and well annotated genome, such as that of Drosophila\[[@B2]\]. Commonly, they are used in time series, for example of development, of life events such as metamorphosis \[[@B3]\], of rhythmic behavior \[[@B4]\] or of responses to environment, such as aging or starvation \[[@B5],[@B6]\]. In Drosophila, arrays are frequently used for whole-organism studies, but in multicellular organisms the ease of experimentation must be balanced against two potential problems: sensitivity and opposing changes. In the first case, even large changes in gene expression in a small tissue will not significantly influence the overall levels in the whole organism; in the second, changes in opposite directions in roughly balanced populations of cells (for example, the sharpening of expression patterns of pair-rule genes) will cancel out at an organismal scale. It is thus vital to resolve gene expression not only over time but also over space. In practice, this means looking at gene expression in defined cell types and tissues as well as in the whole organism. Our assumption is that the expression of many putative genes will go undetected until such tissue-specific studies are performed \[[@B7]\] - with obvious consequences for post-genomics - and we illustrate this point in this paper. We applied Affymetrix arrays in the context of a defined tissue with extensive physiological characterization, the Malpighian (renal) tubule of Drosophila melanogaster. The tubule is a valuable model for studies of both epithelial development and function. Developmentally, the tissue is derived from two distinct origins: an ectodermal outpushing of the hindgut and subsequent invasion (late in embryogenesis) by mesodermal cells \[[@B8]\]. Tubule morphology is very precisely and reproducibly specified; in the tiny tissue of 150 cells, there are altogether six cell types and six regions, specified to single-cell precision \[[@B9]\]. The transport processes that underlie fluid production in the tubule are known in extraordinary detail for so small an organism \[[@B10]-[@B12]\]. The dual origin of the cell types is reflected by dual roles for the ectodermal principal cells and mesodermal stellate cells in the mature tubule; the principal cell is specialized for active transport of cations, whereas the stellate cell appears to control passive shunt conductance \[[@B11],[@B13],[@B14]\]. Cell signaling pathways are also understood in considerable detail: several peptide hormones that act on tubule have been identified \[[@B15]-[@B17]\], and the second messengers cyclic AMP, cyclic GMP, calcium and nitric oxide have all been shown to have distinct roles in each tubule cell type \[[@B10],[@B18]-[@B20]\]. This wealth of physiological knowledge provides a framework for the analysis of the results, and thus - unusually in genetic model organisms - a reality check on the usefulness of the experiment. Results ======= The principle of the experiment was to compare the transcriptome of 7-day adult Drosophila melanogasterMalpighian (renal) tubules, for which defined state there is a wealth of physiological data, with matched whole flies. As described in Materials and methods, data were analyzed by Affymetrix MAS 5.0 software, or by dChip, or dChip and Significance Analysis of Microarrays (SAM) software. Both methods of identifying differentially expressed genes from dChip-normalized data gave virtually the same results. Indeed, SAM analysis followed by further filtering produced 1,465 differentially expressed genes compared to 1,455 genes identified within filtering by dChip alone. Furthermore, the latter list is indeed a subset of the former one. For that reason we report only the list generated by dChip in comparison with MAS data. Both MAS and dChip/SAM gave comparable views of the data, despite the radically different approaches to analysis. It has been shown that the average absolute log ratios between replicate arrays calculated with dChip are significantly lower than one calculated with Affymetrix software (Li and Wong \[[@B21]\]). This bias affecting fold-change calculations is the price of the increased precision that manifests itself in reduced variance, and consequently in the increased sensitivity of identification of differentially expressed genes. Nonetheless, the rank correlation is good (Spearman\'s r= 0.6, p\< 0.0001). Taking genes called as significant by both systems, MAS5 \'up\' call or dChip t-test p-value of 0.01, and narrowing the list by setting an arbitrary cutoff of twofold enrichment and minimum mean difference of 100, MAS5 reported 683 genes and dChip reported 671. Furthermore, the dChip-reported genes overlap with 77% of MAS5-reported genes and this number increases to 91% if only the top 500 MAS5-reported genes are considered. Our confidence in the quality of the dataset is thus high. For simplicity, and because the two analyses produce concordant results, further analysis is restricted to the MAS5 results. The full microarray data have been deposited in ArrayExpress \[[@B22]\]. The fly versus fly and tubule versus tubule samples were extremely consistent, despite the technical difficulty in obtaining the latter (30,000 tubules were dissected in total). In contrast, there was wide divergence between fly and tubule samples (Figure [1](#F1){ref-type="fig"}). Although a common set of housekeeping genes showed comparable abundance, there was a large set of genes enriched in the fly sample, and a smaller set of genes strongly enriched in the tubule sample. In detail, of 13,966 array entries, 6,613 genes were called \'present\' in all five fly samples, compared with 3,873 in tubules. A total of 3,566 genes were present in both fly and tubule: 3,047 in fly only and 307 in tubule only. This illustrates the point that whole-organism views of gene expression are not necessarily helpful in reflecting gene-expression levels in individual tissues. The microarray data are summarized in Tables [1](#T1){ref-type="table"},[2](#T2){ref-type="table"}. Validation of the microarray ---------------------------- Four genes were selected from each of three fly tubule expression classes: very highly enriched; uniformly expressed; and very highly depleted. The expression of each gene was verified by quantitative reverse transcription PCR (RT-PCR) and the data are presented in Table [3](#T3){ref-type="table"}. The agreement between Affymetrix microarray and quantitative PCR determination is good, further increasing our confidence in the robustness of the dataset, and in the approximate correspondence between signal and RNA abundance as a population average. It should be noted that the absolute sizes of the ratios are quite variable; this is a property of dividing a large number by a very small one. Nonetheless, genes scored as enriched or depleted on the arrays are invariably similarly scored by quantitative RT-PCR (QRT-PCR). These data can also be used to validate the use of the normalized Affymetrix signal as a semi-quantitative measure of RNA abundance (Table [1](#T1){ref-type="table"}). If the QRT-PCR dataset of Table [3](#T3){ref-type="table"} is normalized against corresponding signals for rp49(generally taken to be a ubiquitous gene with invariant expression levels in Drosophila), and compared with the globally normalized Affymetrix signal, the agreement is seen to be excellent (Figure [2](#F2){ref-type="fig"}), with a Spearman\'s rof 0.83 (p\< 0.0001). With appropriate caution, the normalized Affymetrix signal can thus be taken as a reasonable estimate of expression levels between genes. Table [1](#T1){ref-type="table"} shows the top 20 genes listed by mean Affymetrix signal intensity. Although this is only a semi-quantitative measure of transcript abundance, the identities of the known genes in the lists are illuminating, and persuade us that the approach has some informal value. Specifically, mRNAs for ribosomal proteins dominate the list, and transporters are conspicuous in the balance. For example, the V-ATPase that energizes transport by tubules is represented by one gene (other subunits are also abundant, but just below the cutoff for Table [1](#T1){ref-type="table"}). The α-subunit of the Na^+^, K^+^ATPase is also highly abundant: this is more surprising, and is discussed below. Two organic cation transporters are also very abundant. Alcohol dehydrogenase, long known to be expressed in tubules \[[@B23],[@B24]\], is also a major transcript. There are also surprises: the most abundant signal is for metallothionein A. This is entirely consistent with our classical understanding of tubule function: it has long been known as a route for metal sequestration and excretion \[[@B25]-[@B30]\]. However, in the entire literature on Malpighian tubules, we are not aware of a physiological investigation of the role of metallothionein, other than documentation of expression \[[@B31],[@B32]\]. The microarray results can thus potently direct and inform future research. Table [2](#T2){ref-type="table"} lists the 53 tubule-enriched genes that are enriched at least 25-fold, in comparison with the whole fly (the full list is provided as an additional data file). The conspicuous feature of these data is the extent to which tubule transcripts differ from any previously published profile. When comparing fly with tubule, there is a large set of genes that are downregulated and another large set of genes that are upregulated in tubule. The extent of the upregulation is also remarkable: the top gene is 99-fold enriched; the top 10 at least 50-fold enriched; and the top 100 at least 16-fold enriched in tubule compared to fly. The standard errors are also extremely low, meaning that we can be very confident (by two separate statistical measures) of the genes called significantly enriched in tubule. The phenotype gap ----------------- Another prominent feature of the signal data in Table [1](#T1){ref-type="table"} is the relatively large fraction of novel genes (those for which there is not even a computer prediction of function) at the top of the list. Indeed, five of the top 10 genes by signal intensity are completely novel - that is, there are no known orthologs - and should provide tantalizing insights into tubule function. The \'phenotype gap\' \[[@B33],[@B34]\] is a key problem in functional genomics; that is, the genetic models preferred for genomics are historically not the organisms selected by physiologists. This can lead to a log-jam in reverse genetics, which depends critically on a wide range of phenotypes to identify effects of the mutation of target genes \[[@B12]\]. It has recently become possible to quantify the phenotype gap \[[@B35]\]. The present dataset elegantly exposes the phenotype gap in Drosophila, and shows that the tubule phenotype may go some way to closing it. Around 20% of Drosophilagenes have been studied in sufficient detail to attract names (beyond the standard \'CG\' notation for computer-annotated genes). Figure [3](#F3){ref-type="fig"} shows that the fraction of anonymous genes in the tubule-enriched list is far higher than would be expected. That is, previous work has tended to overlook these genes. Conversely, because it is possible to perform detailed physiological analysis in tubules, it is possible to close the phenotype gap for these genes. There is a general implication from these data: that functional genomics, in Drosophilaand other species, will rely increasingly on the study of specific tissues, as it is only in this context that expression of genes will be either measurable or explicable. Reconciling array data with function ------------------------------------ Many microarray experiments merely classify enriched genes to their Gene Ontology families. However, the uniquely detailed physiological data available on the Malpighian tubule allows a much more informative approach. The dataset can be validated by inspection, based on known molecular functions in the tissue and new functions can be inferred from abundant or enriched transcripts in the dataset. As the array is relatively comprehensive (corresponding to the 13,500 genes in release 1 of the Gadfly annotation), the results are also relatively authoritative. ### Organic solutes The housekeeping ribosomal transcripts vanish from the enrichment list (Table [2](#T2){ref-type="table"}), which is now dominated by transporters. Intriguingly, these are not for the V-ATPase that is considered to dominate active transport by the tubule, but for organic and inorganic solutes. There is a range of broad-specificity transporters - for organic cations, anions, monocarboxylic acids, amino acids and multivitamins. There are also multiple inorganic anion co-transporters for phosphate and iodide. Most are not only very highly enriched, but also highly abundant. In more detail, the results are remarkable (Table [4](#T4){ref-type="table"}). Nearly every class of transporter is represented, and almost all of these have at least one representative that is both abundant and enriched, implying a very specific renal role; indeed, this table contains the genes with the highest average enrichments of any class, frequently more than 30-fold. Some transporters have been documented implicitly as having a tubule role; many of the classical Drosophilaeye-color mutants also have an effect on tubule color, and have since been shown to encode genes for transport of eye-pigment precursors \[[@B12],[@B36]\]. These genes now turn out to be both abundant and enriched; among the ABC transporters are scarletand white, and among the monocarboxylic acid transporters is CG12286, which we have recently argued to correspond to karmoisin, a probable kynurenine tranporter \[[@B37]\]. Glucose and other sugar transporters are consistently abundant and enriched, implying that sugar transport is a major (and previously unsuspected) role of the tubule. Inorganic transporters are also included in the table; there are also copper and zinc transporters, which is consistent with electron-probe X-ray microanalysis data that heavy metals accumulate in tubule concretions \[[@B38],[@B39]\], and with the extreme abundance of metallothionein A (Table [1](#T1){ref-type="table"}). As well as specific transporters, the tubule is enriched for several families of broad-specificity transporters (organic anion and cation transporters, multivitamin transporters, ABC multidrug transporters and an oligopeptide transporter). When combined these would be capable of excreting a huge majority of organic solutes. These results invite a substantial revision of our interpretation of the role of the tubule. Classically, it is considered to be the tissue that excretes waste material, both metabolites and xenobiotics, and provides the first stage of osmoregulation. However, nearly all work on insect tubules in the last half-century has focused on the ionic basis of fluid secretion and its control, as these are easily measured experimentally. Although there have been sporadic reports on the active transport of organic solutes such as dyes \[[@B40]-[@B42]\], the historical view was of a relatively leaky epithelium, with a paracellular default pathway for those solutes not recognized by specific transporters. While consistent with the more classical view of the tubule, our results also suggest that the insect is emulating a leaky epithelium to produce the primary urine by incorporating a vast array of broad-specificity active transporters in the plasma membranes of what is electrically rather a tight epithelium. Indeed, this interpretation is consistent with other independent data: the intercellular junctions in tubule are known to be of the pleated stellate variety, the invertebrate equivalent of tight junctions \[[@B43]\]; and, like salivary glands, tubule cells are known to be highly polytene \[[@B44]-[@B47]\] or even binucleate \[[@B48]\], adaptations that maximize the size of cells and thus maximize their area/circumference ratios. ### V-ATPases Physiological analysis of the tubule has concentrated on the secretion of primary urine, and the energizing transporter is a plasma membrane proton pump, the V-ATPase \[[@B13],[@B49]-[@B51]\]. This is a large holoenzyme of at least 13 subunits, encoded by 31 Drosophilagenes \[[@B52],[@B53]\]. V-ATPases have two distinct roles, one carried out at low levels in endomembrane compartments of all eukaryotic cells and the other in the plasma membranes of specialized epithelial cells of both insects and vertebrates \[[@B54]\]. In such cells, the V-ATPases can pack the plasma membrane to such an extent that they resemble semi-crystalline arrays when observed by electron microscopy \[[@B55]\]. It is clearly of interest to find out which genes contribute to the plasma-membrane role of the V-ATPase, though this would normally involve difficult and tedious generation of selective antibodies capable of distinguishing between very similar proteins. However, the mRNAs for those V-ATPase subunits enriched in epithelia should also be particularly abundant; one could thus predict that at least one gene encoding each V-ATPase subunit should show enrichment in tubule compared with the rest of the fly. This is indeed the case (Table [5](#T5){ref-type="table"}): invariably, one gene for each subunit is both significantly enriched, and far more abundant, than any other gene encoding that subunit. The reason that the enrichment is not higher is probably because the whole-fly samples contain other epithelia, each with enriched V-ATPase, as minor parts of the overall sample. The array data thus allow a rapid and authoritative prediction to be made on the subunit composition of the plasma membrane V-ATPase. It will be interesting to extend these data to other epithelia in which V-ATPase is known to be functionally significant. ### Na^+^, K^+^- ATPase The role of the classical Na^+^, K^+^-ATPase in tubule is enigmatic. In nearly all animal epithelia, transport is energized by a basolateral Na^+^, K^+^-ATPase, which establishes a sodium gradient that drives secondary transport processes. By contrast, insect epithelia are energized by a proton gradient from the apical V-ATPase \[[@B56],[@B57]\] and, consistent with this, many insect tissues are paradoxically refractory to ouabain, the specific Na^+^, K^+^-ATPase inhibitor \[[@B58]\]. Accordingly, models of insect epithelial function tend not to include the Na^+^, K^+^-ATPase. It is thus interesting to note that both Atpalphaand Nervana 1(encoding isoforms of the α and β subunits, respectively) are among the most abundant transcripts in tubule (Table [6](#T6){ref-type="table"}). Both are about as enriched in tubule as the V-ATPase subunits, but are significantly more abundant (compare Table [5](#T5){ref-type="table"}). By contrast, a novel alpha-like subunit (CG3701), and both Nrv2(the neuronal β-subunit) and other novel β-like subunits are at near-zero levels. As Na^+^, K^+^-ATPase has previously been documented as being particularly abundant in Drosophilatubule \[[@B59]\], it may thus be prudent to re-include the Na^+^, K^+^-ATPase as an important part of models of tubule function. ### Potassium channels Potassium is actively pumped across the tubule, and the main basolateral entry step is via barium-sensitive potassium channels, both in tubule \[[@B50],[@B60],[@B61]\] and in other V-ATPase-driven insect epithelia \[[@B62],[@B63]\]. Of the ion channels, the potassium channel family is by far the most diverse in all animals: in Drosophila, there are at least 28, and in human 255, K^+^-channel genes \[[@B64]\]. Inspection of the potassium channels on the array (Table [7](#T7){ref-type="table"}) clearly identifies just four that are expressed at appreciable levels. Irk3, Ir, Irk2and NCKQare all both very abundant and highly enriched in tubule. Irk3in particular is 80-fold enriched over the rest of the fly, implying a unique role in tubule. Three of these genes are members of the inward rectifier family of potassium channels: supporting the hypothesis that they are critical for potassium entry, these channels are known to be highly barium-sensitive \[[@B65]\]. An inward rectification of potassium current (meaning that potassium would pass much more easily into the cell than out) would be ideal for a basolateral entry step. Inward rectifier channels normally associate with the sulfonylurea receptor (SUR), an ABC transporter, in order to make functional channels \[[@B66],[@B67]\]. In tubules, SURmRNA is present at extremely low abundance (signal 6, enrichment 0.9 times). However, CG9270, a gene with very close similarity to SUR(1 × 10^-28^by BLASTP) is very abundant in tubule (see Table [4](#T4){ref-type="table"}), (signal 422, enrichment 21 times). A second very similar gene, CG31793(previously also known as CG10441and CG17338), is very much less abundant (signal 24, enrichment 0.5). We therefore predict that novel inward rectifiers, formed between Irk3, Iror Ir2and CG9270, may provide the major basolateral K^+^entry path in tubule. In contrast, the other classes of K^+^channel, and the Na/K/Cl co-transporter that has been documented in tubule, are all relatively low in both abundance and enrichment. ### Chloride and water flux In a fluid-secreting epithelium, a necessary correlate of the active transport of cations must be the provision of a shunt pathway for anions and a relatively high permeability to water. In Drosophilatubules, a hormonally regulated chloride conductance pathway has been shown to occur in the stellate cells, although the molecular correlate of the currents has not been determined. There are three ClC-type chloride channels in the Drosophilagenome, and RT-PCR has shown that all three are expressed in tubule \[[@B12]\]. The array data present a prime candidate (Table [8](#T8){ref-type="table"}). Although all three genes are expressed, only one (CG6942) is both very abundant and enriched in tubule (signal 251, enrichment 4). It is thus an obvious candidate partner to provide a shunt pathway for the epithelial V-ATPase. Water flux through the tubule is also phenomenally fast: each cell can clear its own volume of fluid every 10 seconds \[[@B12]\]. Although traditionally it was thought that only a leaky epithelium could sustain such rates, the identification of aquaporins (AQP) (the predominant members of the major intrinsic protein (MIP) family) as major water channels in both animals and plants \[[@B68]\] provides an obvious counter-explanation. There is physiological and molecular data for the presence of aquaporins in Drosophilatubule \[[@B69]\], and AQP-like immunoreactivity has been demonstrated in stellate cells \[[@B12]\]. Table [9](#T9){ref-type="table"} shows that only four of the seven AQP/MIP genes are abundant, and only three enriched. One can thus tentatively assign an organism-wide role to CG7777(signal 243, enrichment 0.6), but tubule-specific roles to CG4019, CG17664and DRIP. In particular, CG17664, is both highly abundant and very highly enriched (signal 705, enrichment 7.9). Control of the tubule --------------------- The hormonal control of fluid secretion is well understood. The major urine-producinig region of the tubule is the main segment \[[@B70]\], and is composed of two major cell types, principal and stellate cells \[[@B9],[@B13],[@B71]\]. Active cation transport in the principal cell is stimulated by the hormones calcitonin-like peptide and corticotrophin releasing factor (CRF)-like peptide, both of which act through cyclic AMP (cAMP). Another peptide family, the CAPA peptides, act through intracellular calcium to stimulate nitric oxide synthase and thus raise cyclic GMP (cGMP), an unusual autocrine role for nitric oxide \[[@B20],[@B72]\]. In the stellate cell, the chloride shunt conductance is activated by leucokinin \[[@B17],[@B73]\], and a role for tyramine as an extracellular signal has also been proposed \[[@B74]\]. So far, the CAPA and leucokinin receptors have been identified \[[@B75],[@B76]\]; both are prominent among the receptors enriched in tubule (Table [10](#T10){ref-type="table"}). The CAPA receptor appears much more highly enriched in tubule than the leucokinin receptor, which is consistent with our understanding of each: the tubule is the only known target of CAPA, whereas leucokinin receptors are widely distributed in the adult gut, gonad and nervous system \[[@B75]\]. There are many other receptors that are reasonably abundant and enriched in tubule. As well as candidate receptors for calcitonin-like and other neuropeptides, there are two glycine/GABA-like receptors that might be expected to form ligand-gated chloride channels, together with good matches to vascular endothelial growth factor-like, insulin-like and bombesin-like receptors. The localization of, ligands for, and functional roles of these receptors will be of great interest. It should be noted in this context that all hormones characterized so far act on one of the two main cell types in the principal section of the tubule. There are, however, six genetically defined cell types and six regions in the adult tubule \[[@B9]\], and it is likely that there will at least be ligands acting on the initial segment to stimulate calcium excretion, and others acting to regulate reabsorption by the lower tubule. If any of these receptors maps to these regions, they would be prime candidates for such roles. Overall, the main surprise from these data is the sheer range of candidate ligands that could be inferred; this more than doubles the size of the endocrine repertoire so far postulated for insect tubules. On a more general level, it is possible to trace out the key genes in all three intracellular signaling pathways that have been studied in detail in Drosophilatubule (Table [11](#T11){ref-type="table"}). The results for signaling genes tend not to be as clear-cut as for transporters, as many are rather widely distributed, and so do not show enrichment, and many do not require high standing levels of protein (and implicitly mRNA) to achieve their effects. Nonetheless, it is possible to identify genes that are at least present, and frequently enriched, in tubule. For the cAMP pathway, it is possible to identify adenylate cyclases, protein kinase A catalytic and regulatory subunits, and a phosphodiesterase (dunce). For cGMP, there are both soluble and membrane guanylate cyclases, implying that the tubules may produce cGMP directly in response to novel ligands, as has recently been suggested \[[@B77]\]. Both Drosophilagenes encoding protein kinase G are expressed in tubule, and one is highly enriched. This is consistent with the renal phenotype observed both in foragingmutants \[[@B78]\], and in tubules in which protein kinase G is overexpressed \[[@B79]\]. There is also a PDE11-like phosphodiesterase. For calcium, two genes for phospholipase C, one for calmodulin, and one for protein kinase C and for calcium/calmodulin-dependent protein kinase are apparent. There are also a number of interesting modulatory or anchoring proteins, such as 14-3-3 zeta, A-kinase anchoring proteins, and receptors for activated C-kinase (Rack1). How is the tubule specified? ---------------------------- The developmental origin of the tubule has been reviewed in detail \[[@B80]-[@B82]\]. Briefly, four unique \'tip cells\', specified by a cascade of neurogenic genes, control cell division in four outpushings (anlagen) of the hindgut, to form the Malpighian tubules. Late in embryogenesis the tubule is invaded by mesodermal cells, which intercalate between the future principal cells, and which then differentiate to form stellate cells \[[@B8]\]. In the adult, there are known to be at least six cell types and six tubule regions \[[@B9]\]. These regions are specified to great precision, and it is clear that each cell in the tubule has a precise positional identity. How does this identity persist throughout the lifetime of the animal? Presumably, combinations of transcription factors interact to provide both regional and cell-type coordinates and, after early establishment, these combinations must persist into adulthood. The microarray data allow the identification of transcription factors that are either highly abundant or highly enriched in tubule. Although this is by no means a complete list of transcription factors that are of importance to tubules, it is a good starting point. Furthermore, there are enhancer trap or reporter gene constructs available for many transcription factors. Accordingly, the top transcription factors and DNA-binding proteins were identified from the array dataset (Table [12](#T12){ref-type="table"}). Some of these transcription factors are already known to be present in tubule, and their presence is confirmed: cut, which is known to be required for development of, and expressed in adult Malpighian tubules \[[@B83]\]; and forkheadand homothorax, both implicated by expression or mutational analysis to be involved in tubule development \[[@B84],[@B85]\]. Teashirt, which has recently been shown to be stellate-cell specific in the late embryo \[[@B8]\], is also present in the adult, with fairly high enrichment (4.6 times). The array results also implicate a further set of transcription factor genes (ETS21C, CG4548, bowl, sequoia, tap, CG1162, pnt, shaven, forkhead domain 59A, sloppy paired 2, lim3) as important in adult. Significantly, these mainly encode transcription factors implicated in development of the nervous system (another ectodermal tissue), so their reuse in the adult tubule is not too surprising. Once the binding sites for these factors are known, it will be interesting to model gene expression in different tubule regions. As transcription factors have been studied experimentally in some detail, they are relatively well represented by enhancer trap and other in vivoconstruct lines. Although individual lines do not necessarily represent the complete expression pattern of their cognate genes, a collection of such lines can provide a rapid first validation of a gene list (Table [12](#T12){ref-type="table"}). Accordingly, representative reporter gene lines were ordered from the Bloomington Stock Center \[[@B86]\], and their adult staining patterns in tubule and gut are shown in Figure [4](#F4){ref-type="fig"}. The results are exciting: most lines showed patterned staining in tubule that is consistent with our original genetically derived map of the tubule \[[@B9]\]. For example, homothoraxmarks out the initial, main and transitional segments of the tubule, whereas CG7417marks the complementary lower tubule domain. The latter line is widely used as a highly specific mushroom body GAL4driver line in brain, and it is interesting that the two known lower tubule GAL4driver lines (c507 and c232) are both insertions in alkaline phosphatase 4, a gene which is only expressed in lower tubule and the ellipsoid bodies of brain (next to the mushroom bodies) \[[@B87]\]. There is also a cell-type-specific transcription factor: cortois found only in stellate cells. Several other transcription factors show ubiquitous, rather than patterned, expression in the tubule, but this is nonetheless consistent with their identification in the microarray dataset. Another interesting aspect of the data in Table [12](#T12){ref-type="table"} is the number of anonymous CG genes implicated in tubule function. These genes have been annotated as transcription factors because of DNA-binding domains, for example, but have not been characterized functionally. The epithelial phenotype gap is thus evident even in this most intensely studied group of genes. Exceptions to the rule ---------------------- The whole premise of microarray work is that an abundant or enriched signal indicates the importance of a gene product in a particular context. This hypothesis is normally both untested and unchallenged. The unusual depth of functional understanding of the tubule allows a more rigorous appraisal. In fact, the majority of the genes implicated in tubule function are found well up the list. There are, however, several conspicuous exceptions (Table [13](#T13){ref-type="table"}). The calcium channels trpand trplare normally considered to be eye-specific, and have an essential role in phototransduction \[[@B88]-[@B90]\]. It is thus not surprising to find both genes almost at the bottom of the gene list. We have shown, however, that fluid secretion is severely compromised by mutations in either gene. Similarly, nitric oxide synthase (NOS) is a major signal transducer in tubule \[[@B20],[@B72]\]. Nonetheless, all three genes are within the \'bottom\' 20 of the whole array, with signals that are barely detectable and significant depletion compared with the whole fly. This is a cautionary example: while abundant or enriched signals can be taken as reliable indicators of functional significance, the converse is not necessarily true. The tubule and human disease ---------------------------- Consequent to the demonstration of the phenotype gap, there are some intriguing, abundant and enriched genes which by virtue of their non-uniform expression, are likely to be important in (and best studied in) tubule. A systematic approach was taken by combining the tubule-enriched gene list with the homophila database of Drosophilagenes with known human disease homologs. The results (Table [14](#T14){ref-type="table"}) show the 50 human diseases with Drosophilahomologs that are upregulated at least threefold in tubules. Intriguingly, several of these genes have human kidney phenotypes. Some are extremely well studied: for example, rosy(one of the first Drosophilamutations recorded) encodes xanthine oxidase, and mutation in either human or fly produces severe nephrolithiasis with concomitant distortion of tubules (reviewed in \[[@B12]\]). The distension of tubules is remarkable (Figure [5](#F5){ref-type="fig"}). In both species, lethal effects can be ameliorated by a high-water, low-purine diet. Other diseases, although less well documented, have plausible renal phenotypes: for example, antenatal Bartter syndrome, a severe salt-wasting renal disease, associated with mutations in the ROMK channel (homolog ir); Dent disease, caused by mutation in ClC5(homolog CG5284); proximal renal tubular acidosis, caused by mutation in the NDAE co-transport (homolog ndae1); nephrophatic cystinosis, caused by mutation in a lysosomal cystine transporter (homolog CG17119); mucopolysaccharidosis type IV, caused by mutation in galactosamine-6-sulphatase, an enzyme enriched in both human and fly kidney (homolog CG7402). Overall, there is a clear message that human and fly renal function may be relatively similar over quite a wide range of properties. The tubule phenotype may also prove highly informative for other genes implicated in disease. Recently, a small 10 kDa protein, bc10, was shown to be downregulated in the transition from early-stage to invasive bladder carcinoma \[[@B91]\]. The normal function of this protein is not yet established, but its homolog (bc10) is highly abundant (893 ± 50) and moderately enriched (1.9 ± 0.09) in tubule, and a P-element insertion within the gene P{GT1}BG02443, is available from stock centers. This comparative approach can be extended to non-human species. For example, CG4928represents an abundant and enriched transcript (3,778, 13 times enriched), that is highly similar (1.9 × 10^-75^) to the C. elegansgene unc-93\[[@B92]\]. This is associated with a \'rubber-band\' phenotype, in which motor co-ordination is sluggish; it is thus taken to be a myogenic or neuromuscular gene. The discovery that a close homolog is highly enriched in renal tissue opens new lines of investigation for this gene. Discussion ========== These data have value at two distinct levels: specific and general. Specifically, we have found out more about the operation of the Malpighian tubule than in any single published piece of work since the very first pioneering days: a summary is given in Figure [6](#F6){ref-type="fig"}. This tissue is of great interest, both for developmental studies and for integrative physiological study of epithelial function. Despite 990 papers on Malpighian tubules since the start of the twentieth century, and a really rather good understanding of ion and water transport, the microarray data provide strong indications that these are only minor properties of the tubule. Whole families of transporters are represented by abundant mRNAs and transport solutes that have yet to be studied in the context of tubule. Some datasets implicate particular genes in processes that have been studied in great physiological detail, and the presence of known genes with the novel can only increase our confidence in the result. In this context, the demonstrated abundance of transporters for almost every class of organic and inorganic solute dramatically diminishes the number of solutes for which a nonspecific paracellular pathway need be invoked. The data thus allow the conceptual view of the epithelium to alter from leaky to tight in a physiological-transport sense: this is consistent with electrophysiological data \[[@B93]\]. There are two areas where microarray data deserve comment. Firstly, more than 300 genes are expressed in tubule but called as absent in whole-fly samples. Although there is an obvious convenience and consistency in employing whole-organism samples for array studies, it is important to recognize that the approach is very likely to suppress the detection of those interesting genes that are not widely expressed. Secondly, the premise that abundance on an array (or more generally, abundance of an RNA species) necessarily correlates with functional significance can be spectacularly refuted by three examples, the trpand trplchannels and NOS. It is, however, probably significant that these are cell-signaling molecules, where a relatively small number of molecules can have a disproportionate influence on cell behavior. By contrast, the transport genes for which the tubule is so enriched are much more likely to exert effects proportional to their abundance. Conclusions =========== Reverse genetics is a vital tool in functional genomics, but the \'phenotype gap\' has hampered widespread implementation of this approach \[[@B35]\]. As the tubule presents a range of easily assayed phenotypes \[[@B12]\], this work specifically identifies those genes that are likely to be best studied in tubule by virtue of their very high enrichment. In addition to the obvious transport genes, it is interesting that many transcription factors and human disease gene homologs fall into this category. This work thus stresses the importance of systematic, fine-grained, tissue-specific microarray analysis in closing the phenotype gap for multicellular model organisms. Materials and methods ===================== Flies ----- Drosophila melanogasterwere kept on standard diet at 25°C and 55% relative humidity on a 12:12 h photoperiod. Malpighian tubules were dissected from 7-day-old adults, for compatibility with the extensive physiological literature on the tubule \[[@B10],[@B11],[@B13],[@B15],[@B17],[@B19],[@B20],[@B39],[@B70],[@B75],[@B94]-[@B96]\]. At this stage, the tubules are in a relatively stable state after adult emergence, and their secretion parameters do not change detectably between 3 and 14 days post-emergence. Microarrays ----------- Tubules were dissected in batches of 1,000 by a group of eight experimenters. Tubules were aggregated into Trizol every 15 min to minimize the distortion of the transcriptome by the trauma of dissection and in vitroincubation. Care was taken to sever the tubules from the gut at the lower ureter so that no other tissue was included in the sample. For each experimental point, whole flies from the same culture were homogenized in Trizol in batches of 100, to permit a matched pair comparison. Six repeats were performed. RNA was extracted according to standard protocols, and quality was assessed with an Agilent RNA Bioanalyzer. Samples of 20 μg total RNA were reverse-transcribed, then in vitrotranscribed, according to Affymetrix standard protocols. The quality of the ccomplementary RNA (cRNA) was also checked on an Agilent RNA Bioanalyzer, with a sample in which the broad cRNA peak exceeded the height of the low molecular weight degradation peak taken to be satisfactory. Samples were then run on the Affymetrix Drosophilagenome array under standard conditions. Quality control was at several levels: the Affymetrix MAS 5.0 software provided evidence of successful sample preparation, with test genes providing a 3\':5\' signal ratio of less than 3. dChip \[[@B97]\] provided an alternative view, with a direct oligo-by-oligo view on the success of hybridization across the array surface; slides with both single-probe and probe-set outlier rates of less than 5% were taken as satisfactory. Only arrays in which both results were in range were accepted. In this case, 11 of 12 arrays were satisfactory; the first tubule array failed both MAS and dChip criteria, and so the first experimental pair was discarded to leave a five-sample paired design. As will be seen from the results, this design was sufficient to identify tubule-enriched genes with a high level of confidence. As sample collection extended over the whole day, array results from morning versus afternoon samples were compared (data not shown), but no difference was found between the two groups at this very broad time resolution. Bioinformatics -------------- Microarray samples were analyzed by two independent routes. The first was low-level analysis with the Affymetrix MAS 5.0 suite and identification of differentially expressed genes using the Affymetrix Data Mining Tool. The second was low-level analysis using dChip software \[[@B97]\] followed by assessment of significance using SAM software \[[@B98]\] followed by post-analysis by dChip. The MAS5 low-level analysis consisted of background subtraction followed by robust conversion of probe-level perfect match-mismatch (PM-MM) expression values into probe-set-level signals followed by linear multi-chip normalization (scaling). Tubule enrichment was based on an Affymetrix \'up\' call, and a critical level of p\< 0.05. In this analysis method, tubule and fly samples were taken as matched pairs, reflecting their biological origin. The dChip-based low-level analysis consisted of background correction followed by the multi-chip, \'invariant-set\' nonlinear normalization at probe level followed by the calculation of model-based expression indices using PM expression values only. Differentially expressed genes between two groups of five replicates were identified within dChip by filtering data using the following criteria: lower 90% confidence bound of fold-change \[[@B21]\] \> 2; difference between group means on antilog scale \> 100 and p-value for t-test of equal group means \< 0.01. Alternatively, the differentially expressed genes were identified using SAM software with 1,000 sample permutations and false-discovery rate cutoff of 1%. These were then post-filtered using two first criteria from the dChip analysis mentioned above. Fold change was calculated as a ratio of group means. Outputs were saved as Excel files, and parsed by hand-coded Perl scripts. Additional data file ==================== A list of genes (Additional data file [1](#s1){ref-type="supplementary-material"}) called as upregulated in tubule by Affymetrix SAM 5 software, and with more than two-fold enrichment is available with the online version of this article. Supplementary Material ====================== ::: {.caption} ###### Additional data file 1 A list of genes called as upregulated in tubule by Affymetrix SAM 5 software, and with more than two-fold enrichment ::: ::: {.caption} ###### Click here for additional data file ::: Acknowledgements ================ We thank the staff of the Sir Henry Wellcome Functional Genomics facility in Glasgow, for their help and training in Affymetrix technology. We thank the following members of the Dow/Davies lab for their assistance in the \'ten thousand tubule days\': Laura Kean, Valerie Pollock, Shirley Graham, Kate Broderick, Matthew Macpherson, Kostas Stergiopoulos and Pablo Cabrero. This work was funded by BBBSRC GAIN grants to J.A.T.D and S.A.D. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Scatterplot of mean whole fly vs tubule signal intensities. Genes called as significantly enriched in tubule compared with fly by MAS 5.0 are in red, those significantly depleted in blue, and those not significantly different in yellow. ::: ![](gb-2004-5-9-r69-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Semi-quantitative inter-gene comparison is possible using Affymetrix signal. The 24 QRT-PCR results underlying Table 3 were normalized against rp49, and plotted against the Affymetrix signal globally normalized as in MAS 5.0. Spearman\'s rwas calculated, and significance of the correlation assessed (one-tailed), using Graphpad Prism 3.0. ::: ![](gb-2004-5-9-r69-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### The phenotype gap. Genes enriched in tubules are historically under-researched. The percentage of genes with explicit names (other than automatic CG annotations) is shown for the entire genome, and for the top 50, 100 and 200 genes (as judged by fold enrichment) from the tubule dataset. ::: ![](gb-2004-5-9-r69-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Expression patterns in tubules of some of the transcription factor genes indicated by the microarray data as being expressed in tubules. (a)homothorax(hth^05745^), principal and stellate cells of initial and transitional segments only; (b)polyhomeotic proximal(ph-p), all cells of tubule, and midgut; (c)pointed(pnt^1277^), principal and stellate cells of initial and transitional segments only; (d)corto(corto^07128b^), stellate cells only; (e)teashirt(tsh^04319^, a kind gift of H. Skaer), stellate cells only; (f)bunched(bnc^00255^), principal cells, whole tubule; (g)cut(immunocytochemistry, antibody a kind gift of Jan lab), whole tubule, principal cells only; (h)CG7417(CG7417^201Y^), lower tubule (and midgut - strong); (i)arc(a^k11011b^), lower tubule, not ureter; (j)Stat92E^06346^, all tubule cells and midgut. ::: ![](gb-2004-5-9-r69-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Recapitulation of human xanthinuria type 1 by rosymutants. (a)Wild-type tubule; (b)tubule from adult ry2homozygous fly. Both micrographs are at the same magnification, and the diameter of the wild-type tubule can be taken as 35 μm. ::: ![](gb-2004-5-9-r69-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Summary of major genes enriched in tubule. Genes shown are upregulated at least threefold. ::: ![](gb-2004-5-9-r69-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Most abundant genes in tubule, sorted by normalized Affymetrix signal strength ::: Gene Signal Enrichment Function ----------------- -------------- ------------ ------------------------------------------- MtnA 12,114 ± 581 3.0 ± 0.0 Cu-binding CG7874 10,672 ± 518 7.4 ± 0.4 CG14292 10,392 ± 572 8.4 ± 0.5 CG3168 10,199 ± 459 6.2 ± 0.3 Transporter RpS25 9,368 ± 276 1.3 ± 0.0 Small-subunit cytosol ribosomal protein Adh 8,895 ± 395 1.3 ± 0.0 Alcohol dehydrogenase; EC 1.1.1.1 RpS20 8,720 ± 226 1.2 ± 0.0 Small-subunit cytosol ribosomal protein CG13315 7,818 ± 370 3.9 ± 0.6 CG14245 7,767 ± 305 13.4 ± 2.3 RpL27A 7,757 ± 198 1.3 ± 0.0 Large-subunit cytosol ribosomal protein CG18282 7,711 ± 160 1.7 ± 0.0 RpL18A 7,514 ± 200 1.4 ± 0.0 Large-subunit cytosol ribosomal protein RpL14 7,483 ± 209 1.3 ± 0.0 Large-subunit cytosol ribosomal protein RpP2 7,481 ± 283 1.3 ± 0.1 Cytosolic ribosomal protein CG6726 7,307 ± 244 14.4 ± 0.5 Peptidase RpL23a 7,284 ± 254 1.2 ± 0.1 Large-subunit cytosol ribosomal protein CG4046 7,250 ± 165 1.1 ± 0.1 Structural protein of ribosome CG7084 7,211 ± 329 36.8 ± 6.5 Transporter RpL3 7,179 ± 105 1.4 ± 0.1 Large-subunit cytosol ribosomal protein CG9914 7,088 ± 466 12.0 ± 1.4 Enzyme CG3203 7,024 ± 219 1.3 ± 0.1 L17-like CG6846 6,989 ± 177 1.3 ± 0.1 Structural protein of ribosome blw 6,890 ± 142 1.7 ± 0.0 ATP synthase alpha subunit BcDNA:GH08860 6,742 ± 278 5.0 ± 0.3 Enzyme RpS3 6,709 ± 240 1.3 ± 0.1 DNA-(apurinic or apyrimidinic site) lyase CG5827 6,603 ± 169 1.3 ± 0.1 Structural protein of ribosome CG15697 6,543 ± 174 1.3 ± 0.1 Structural protein of ribosome RpS9 6,502 ± 171 1.2 ± 0.0 Small-subunit cytosol ribosomal protein Rack1 6,463 ± 105 1.3 ± 0.0 Protein kinase C binding protein vha26 6,416 ± 190 3.1 ± 0.3 V-ATPase E subunit Ser99Da 6,305 ± 2100 0.6 ± 0.2 Serine carboxypeptidase Ser99Db 6,300 ± 2119 0.6 ± 0.2 Serine-type endopeptidase CG1883 6,258 ± 172 1.2 ± 0.1 Structural protein of ribosome RpL32 6,251 ± 217 1.3 ± 0.1 Large-subunit cytosol ribosomal protein Atpalpha 6,240 ± 151 4.2 ± 0.1 Na, K-ATPase alpha subunit CG3270 6,234 ± 167 32.3 ± 2.6 Sarcosine oxidase RpS26 6,080 ± 151 1.3 ± 0.1 Small-subunit cytosol ribosomal protein sop 6,070 ± 157 1.1 ± 0.0 Small-subunit cytosol ribosomal protein RpL7 6,060 ± 113 1.2 ± 0.0 Large-subunit cytosol ribosomal protein CG3321 6,034 ± 122 1.6 ± 0.0 Enzyme eIF-4a 6,027 ± 270 1.9 ± 0.1 CG8857 5,977 ± 309 1.4 ± 0.1 Structural protein of ribosome oho23B 5,940 ± 176 1.3 ± 0.1 Ribosomal protein CG3762 5,874 ± 79 4.2 ± 0.1 CG9091 5,850 ± 281 1.2 ± 0.1 Structural protein of ribosome vha16 5,845 ± 215 2.6 ± 0.1 V-ATPase c subunit CG18323 5,820 ± 201 1.5 ± 0.1 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Genes enriched more than 25-fold in tubules ::: Gene Product MAS enrichment ----------- ------------------------------------------------ ---------------- CG13365 98.9 CG14957 95.9 CG13905 85.2 CG13836 80.6 Irk3 Potassium channel protein-like 80.3 CG14963 55.4 CG3014 54.0 CG13161 53.8 CG17043 49.9 CG18095 47.8 CG13656 45.5 CG13311 43.5 CG17817 40.9 CG9434 40.6 CG17522 Glutathione transferase 39.5 CG15359 38.7 CG7084 Organic cation transporter 36.8 CG8028 Monocarboxylate transporter-like 36.6 CG8951 Sodium-dependent multivitamin transporter-like 35.8 CG3690 34.8 CG15406 Sugar transporter 34.5 CG14293 33.5 CG17028 33.4 CG3285 Sugar transporter-like 33.0 CG3270 32.3 scarlet ATP-binding cassette (ABC) transporter 32.3 CG6529 Sugar transporter-like 32.1 CG2680 4-nitrophenylphosphatase-like 31.2 CG8620 30.5 CG15279 Cation amino-acid symporter 30.1 CG9509 29.7 CG14539 29.3 CG3382 Organic anion transporter 29.3 CG6602 29.3 CG5361 Alkaline phosphatase-like 29.2 CG8957 Iodide symporter-like 29.1 CG10006 29.0 CG15155 28.9 CG10226 ATP-binding cassette transporter 28.3 CG2196 Sodium iodide symporter 27.7 CG16762 27.6 CG14195 27.4 CG8125 Aryldialkylphosphatase 27.4 CG7881 Sodium phosphate cotransporter 27.1 CG8934 Sodium iodide symporter-like 27.1 CG7402 N-acetylgalactosamine-4-sulfatase-like 26.9 NaPi-T Na phosphate cotransporter 26.8 CG8791 Sodium phosphate cotransporter 26.8 CG8776 Cytochrome b561-like 26.6 CG3212 26.6 CG14857 Organic cation transporter-like 26.4 CG8932 Sodium-dependent multivitamin transporter-like 25.9 Cyp6a18 Cytochrome P450, CYP6A18 25.5 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Validation of array data by QRT-PCR ::: Gene MAS enrichment SAM enrichment QRT-PCR enrichment ------------------ ---------------- ---------------- -------------------- Highly enriched CG13665 98.9 8.7 9.0 CG14957 95.9 21.9 23.8 CG13905 22.6 17.4 110 CG13836 80.6 30.1 11.7 Evenly expressed CG17737 1.0 0.9 0.74 CG10731 1.0 1.1 0.68 CG8327 1.0 0.8 1.2 Arp66 1.0 1.1 0.47 Highly depleted CG13421 0.00 0.067 0.19 CG12408 0.01 0.11 0.14 Act88F 0.01 0.14 0.03 CG15575 0.01 0.082 0.008 Enrichment in tubule mRNA compared to whole fly mRNA, computed from the microarray dataset with MAS 5.0 or SAM (see text), were compared with real values obtained by QRT-PCR. Four separate fly and tubule samples were run with primers for each gene, and for rp49, a ribosomal gene generally considered to be invariant. RNA quantities were calculated, and the gene:rp49ratio calculated for each sample pair. Tubule enrichment was calculated as the (gene:rp49)~tubule~/(gene:rp49)~fly~. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Transporters sorted by class ::: Gene/class Signal Enrichment ------------------------------------------------------------- -------------- ------------ ATP-binding cassette (ABC) transporter (6/46) st 1,521 ± 34 32 ± 2.8 CG10226 290 ± 25 28 ± 3.4 CG9270 422 ± 21 21 ± 2.7 w 798 ± 53 10 ± 1.4 bw 18 ± 2 4 ± 1.2 CG17338 72 ± 6 3 ± 0.2 Cationic amino-acid transporter (1/5) CG7255 308 ± 34 7 ± 0.8 Copper transporter (1/6) CG7459 374 ± 6 5 ± 0.6 Monocarboxylate transporter (4/14) CG8028 2,567 ± 82 37 ± 2.1 CG8468 1,377 ± 67 10 ± 0.7 CG8389 698 ± 38 4 ± 0.2 CG12286(kar) 550 ± 15 3 ± 0.1 Multidrug efflux transporter (1/6) CG8054(now CG30344) 1,366 ± 68 6 ± 0.4 Pyrimidine-sugar transporter of Golgi (1/1) CG3874(frc) 877 ± 40 5 ± 0.3 Oligopeptide transporter (1/3) CG9444 517 ± 12 10 ± 1.2 Organic anion transporter (3/5) CG3382 1,076 ± 56 29 ± 3.3 CG3380 3,385 ± 126 24 ± 1.6 CG6417 678 ± 90 9 ± 2.4 Organic cation transporter (11/21) CG7084 7,211 ± 329 37 ± 6.5 CG14857 472 ± 13 26 ± 5.5 CG17751 1,331 ± 34 25 ± 4.2 CG16727 3,152 ± 200 23 ± 3.2 CG17752 4,847 ± 37 21 ± 2.1 CG14856 36 ± 5 7 ± 2.4 CG3168 10,199 ± 459 6 ± 0.3 CG6231 269 ± 30 5 ± 1.0 CG7342 20 ± 2 5 ± 1.5 CG8654 274 ± 29 4 ± 0.6 Reduced folate transporter (2/3) CG14694 584 ± 22 13 ± 1.6 CG6574 190 ± 8 4 ± 0.3 Sodium bicarbonate cotransporter (1/1) CG4675(Ndae1) 531 ± 34 5 ± 0.5 Sodium-dependent inorganic phosphate cotransporter (1 / 20) NaPi-T 1,430 ± 428 27 ± 2.3 Sodium-dependent multivitamin transporter (4/5) CG8951(now CG31090) 1,363 ± 30 36 ± 3.9 CG8932 2,106 ± 130 26 ± 1.4 CG8451 365 ± 10 4 ± 0.4 CG10879(now CG31668) 6 ± 1 3 ± 0.7 Glucose transporter (3/17) CG7882 4,951 ± 171 16 ± 0.8 CG8249 302 ± 12 6 ± 1.0 Glut1 342 ± 24 3 ± 0.2 Sugar transporter (7/7) CG15406 5,322 ± 186 35 ± 2.8 CG3285 1,405 ± 55 33 ± 1.3 CG6529(now CG31272) 3,774 ± 131 32 ± 4.8 CG15407 840 ± 44 25 ± 2.1 CG14606 1,210 ± 56 22 ± 2.0 CG15408 3,333 ± 194 21 ± 1.7 CG8837 1,277 ± 88 19 ± 2.9 Zinc transporter (4/6) BG:DS07295.1(now CG3994) 3,608 ± 91 10 ± 1.0 CG4334 378 ± 19 5 ± 0.4 CG17723 919 ± 59 4 ± 0.3 CG5130 104 ± 10 4 ± 0.6 For brevity, only family members enriched by more than threefold are shown. For each grouping, the numbers in parentheses refer to the number of genes enriched in tubule, compared to the total number of such genes in the Drosophilagenome, as classified by Gene Ontology. Where original gene names have been superseded by later annotations of the Drosophilagenes, the new names are shown in parentheses. ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### V-ATPase genes that are enriched in tubule ::: Subunit Copy number Genes Affymetrix reference Signal Enrichment ------------ ------------- ------------------------------------- ---------------------- ----------------- --------------- V~1~sector A 3 vha68-1(CG12403) 142380\_at 9 ± 2 0.5 ± 0.1 vha68-2(CG3762) 146305\_at 5,874 ± 79 4.2 ± 0.1 vha68-3(CG5075) 146306\_at 2 ± 0 0.04 ± 0.02 B 1 vha55(CG17369) 153041\_at 2,304 ± 74 2.7 ± 0.1 SFD (H) 1 vhaSFD(CG17332) 144191\_at 2,671 ± 66 4.4 ± 0.2 C 1 vha44(CG8048) 153422\_at 1,400 ± 74 3.5 ± 0.1 D 3 vha36-1(CG8186) 152480\_at 2,846 ± 154 4.5 ± 0.4 vha36-2([CG13167]{.underline}) 147073\_at 2 ± 0.4 0.1 ± 0.0 [CG8310]{.underline} 144407\_at 29 ± 4 0.6 ± 0.09 E 1 vha26(CG1088) 151930\_at 6,416 ± 190 3.1 ± 0.3 F 2 vha14-1(CG8210) 143625\_at 3,722 ± 105 3.2 ± 0.2 vha14-2(CG1076) 149368\_at 5.6 ± 1.6 1.5 ± 1.1 G 1 vha13([CG6213]{.underline}) 144156\_at 2,952 ± 68 3.3 ± 0.1 V~0~sector A 5 vha100-1(CG1709) 153997\_at 155 ± 8 0.8 ± 0.0 vha100-2(CG7679, CG18617) 142661\_at 3,718 ± 157 5.4 ± 0.3 vha100-3(CG30329) not on array [CG12602]{.underline} 146249\_at 306 ± 26 1.3 ± 0.1 vha100-4([CG7678]{.underline}) 141662\_at 66 ± 3 0.24 ± 0.04 c 5 vha16(CG3161) 141528\_at 5,845 ± 215 2.6 ± 0.1 vha16-2(CG32089)/vha16-3 148578\_at 32 ± 7 1.4 ± 0.22 (CG32090) vha16-4(CG9013) 147341\_at 18 ± 4 1.4 ± 0.6 vha16-5([CG6737]{.underline}) 146189\_at 36 ± 7 0.6 ± 0.12 PPA1 2 vhaPPA1-1(CG7007) 142158\_at 1,895 ± 79 4.1 ± 0.2 (c\") vhaPPA1-2(CG7026) 149926\_at 57 ± 7 0.9 ± 0.1 M9.7 3 vhaM9.7-1(CG11589) 154011\_at 101 ± 9 1.8 ± 0.0 (e, H) [CG1268]{.underline} 148161\_at 14 ± 1 0.1±0.0 vhaM9.7-2(CG7625) 149187\_at 3,101 ± 127 2.9 ± 0.1 AC39 2 vhaAC39-1(CG2934) 154279\_at 2,082 ± 52 3.4 ± 0.1 (d) vhaAC39-2([CG4624]{.underline}) 150428\_at 13 ± 2 0.8 ± 0.12 All genes significantly similar to known human or yeast V-ATPase subunits were identified by BLAST search, extending our previously reported annotation of the V-ATPase family \[53\], by identifying the genes underlined above as V-ATPase subunits. For comparison, enrichment ratios significantly greater than 1 and signals over 1,000 are shown in bold. (vha16-2and vha16-3are in tandem repeat and share the same Affymetrix oligo set, and so cannot be distinguished here.) ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Na^+^, K^+^-ATPase ::: Gene Signal Enrichment ------------ ------------- ------------- α-subunit Atpalpha 6,240 ± 151 4.22 ± 0.05 CG3701 6 ± 1 0.85 ± 0.17 β-subunit Nrv1 1,924 ± 71 3.47 ± 0.21 Nrv2 2 ± 1 0.09 ± 0.06 CG11703 7 ± 2 0.46 ± 0.18 CG5250 4 ± 0 0.18 ± 0.04 CG8663 20 ± 1 0.1 ± 0.01 Although the DrosophilaNa^+^, K^+^-ATPase has classically been thought to be composed of a dimer of Atpalphaand either Nrv1or Nrv2, the other genes here are more similar by BLASTX to the corresponding alpha and beta subunits than any other gene (data not shown). They are thus included in the table as candidate alternative subunits. ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Potassium channels and symporters ::: Gene Signal Enrichment ----------------------------------------------- ------------ -------------- Potassium channels Irk3 (CG10369) 2771 ± 145 80.31 ± 7.75 Ir (CG6747) 1302 ± 112 14.19 ± 1.58 Irk2 (CG4370) 527 ± 33 5.69 ± 0.24 KCNQ (CG12215) 101 ± 0 6.44 ± 2.31 KCNQ (CG12915) 111 ± 10 2.84 ± 0.46 CG10864 29 ± 7 3.74 ± 1.12 CG32770 (CG6952) 5 ± 2 2.6 ± 1.05 elk 5 ± 3 2.23 ± 1.19 CG9361 6 ± 2 2.31 ± 0.84 CG12214 101 ± 11 2.15 ± 0.51 CG7640 12 ± 5 1.48 ± 0.76 eag 8 ± 1 1.59 ± 0.39 Shaker cognate b 6 ± 1 1.38 ± 0.54 CG4450 4 ± 0 1.62 ± 0.28 Shaw 26 ± 4 1.21 ± 0.54 CG1756 15 ± 4 1.31 ± 0.35 Shaker 26 ± 4 1.42 ± 0.23 CG9637 3 ± 1 1.32 ± 0.22 Shal 29 ± 4 1.19 ± 0.29 CG3367 6 ± 1 1.09 ± 0.09 CG8713 41 ± 3 0.9 ± 0.1 Sh 7 ± 3 0.65 ± 0.25 CG9194 8 ± 1 0.77 ± 0.12 CG15655 13 ± 3 0.45 ± 0.13 Ork1 28 ± 2 0.32 ± 0.03 sei 10 ± 2 0.21 ± 0.04 Shab 6 ± 1 0.21 ± 0.04 CG12904 4 ± 1 0.14 ± 0.07 CG17860 5 ± 2 0.1 ± 0.04 Hk 4 ± 1 0.07 ± 0.01 Calcium-activated potassium channels CG10706 21 ± 5 1.93 ± 1.06 slo 2 ± 0 0.11 ± 0.02 CG4179 4 ± 1 1.55 ± 0.91 Potassium-dependent sodium-calcium exchangers CG14744 39 ± 1 1.32 ± 0.12 CG1090 35 ± 2 0.81 ± 0.13 CG14743 5 ± 1 0.48 ± 0.19 Nckx30C 31 ± 5 0.38 ± 0.05 CG12376 8 ± 2 0.24 ± 0.07 Nckx30C 31 ± 5 0.11 ± 0.05 Sodium/potassium/chloride symporter EG:8D8.3 132 ± 6 2.46 ± 0.36 CG10413 185 ± 25 1.75 ± 0.22 CG5594 65 ± 5 0.8 ± 0.07 CG2509 60 ± 6 0.42 ± 0.06 CG4357 12 ± 4 0.12 ± 0.04 ::: ::: {#T8 .table-wrap} Table 8 ::: {.caption} ###### Chloride channels ::: Gene Signal Enrichment ---------- --------- ------------- CG6942 251 ± 9 4 ± 0.29 CG8594 57 ± 5 0.86 ± 0.09 CG5284 100 ± 5 2.2 ± 0.16 These are the three genes with clear similarity to the ClC gene family of vertebrates \[12\]. ::: ::: {#T9 .table-wrap} Table 9 ::: {.caption} ###### Aquaporins and other major intrinsic proteins ::: Gene Signal Enrichment ----------------- ------------ ------------ CG4019 1666 ± 167 2.7 ± 0.3 CG17664 705 ± 91 7.9 ± 0.9 DRIP 318 ± 16 3.6 ± 0.4 CG7777 243 ± 11 0.6 ± 0.06 CG12251 (AQP) 22 ± 3 0.5 ± 0.04 CG5398 8 ± 1 0.2 ± 0.05 bib 2 ± 1 1.1 ± 0.3 ::: ::: {#T10 .table-wrap} Table 10 ::: {.caption} ###### Receptors called as upregulated in tubule, with enrichments more than threefold ::: Gene Signal Enrichment ---------------------------------------------------------- ---------- ------------ CG3212 85 ± 11 27 ± 11 CG17415(calcitonin-like) 633 ± 48 17 ± 2 CG17084 288 ± 27 14 ± 2 CG1147(neuropeptide Y-like) 34 ± 2 13 ± 8 CG14575(CapaR) 311 ± 24 11 ± 1 CG7431(octopamine-like) 40 ± 4 8.5 ± 0.9 CG12414 (nAChRalpha @ 80B) 9 ± 3 8 ± 3.6 CG7589(ligand-gated Cl channel) 564 ± 35 7 ± 0.9 CG12370(diuretic hormone-like) 203 ± 17 6.7 ± 9 CG15556 221 ± 12 6.4 ± 0.5 CG11340(glycine-gated channel-like) 143 ± 8 5.0 ± 0.9 CG14593(bombesin) 59 ± 13 5 ± 2 CG6390(insulin-like growth factor) 85 ± 8 4.3 ± 0.6 CG8222(Pvr, vascular endothelial growth factor-like) 294 ± 26 4.2 ± 0.5 CG6536 42 ± 5 4 ± 1.7 nAcRalpha 24 ± 4 4 ± 1.5 CG7404(steroid-like) 239 ± 21 3.5 ± 0.4 CG10626(LkR) 142 ± 7 2.9 ± 0.4 ::: ::: {#T11 .table-wrap} Table 11 ::: {.caption} ###### Major genes of the cAMP, cGMP and calcium signaling pathways ::: Function Gene name Signal Enrichment Comments ------------------------------------------------------------------------------------------------- -------------------- ------------- ------------- ---------------------------------------- cAMP Adenylate cyclase rutabage 121 ± 12 1.4 ± .2 Ac78C 44 ± 5 7.2 ± 1.6 Ac13E 106 ± 4 4.1 ± 0.5 Protein kinase A Pka-C3 88 ± 9 1.7 ± 0.2 Catalytic subunit Pka-R1 183 ± 13 1.2 ± 0.1 Regulatory subunit PDE dunce 147 ± 6 3.9 ± 0.6 cAMP-specific Calcium CamKinase Caki 112 ± 10 1.7 ± 0.2 Phospholipase C Small wing 46 ± 6 1.1 ± 0.2 Plc21C 58 ± 5 1.1 ± 0.1 Calcium release channels Itp-r83A 11 ± 2 1.2 ± 0.2 InsP~3~receptor Calmodulin Calmodulin 1,019 ± 57 0.9 ± 0.06 Protein kinase C Pkc98E 217 ± 15 1.7 ± 0.2 cGMP Guanylate cyclase CG14885 13 ± 3 6 ± 2.5 Probably soluble beta subunit Gyc76C 410 ± 23 2.9 ± 0.4 Membrane form CG4224 23 ± 4 0.8 ± 0.2 Membrane form CG9873 137 ± 5 2.0 ± 0.7 Membrane form Gycbeta100B 20 ± 4 0.8 ± 0.1 Cytoplasmic, beta subunit CG5719 9 ± 3 3.5 ± 1.4 Membrane form PDE CG10231 182 ± 4 3.7 ± 0.6 cGMP-specific, PDE11-like Protein kinase G foraging 91 ± 2 0.3 ± 0.01 Pkg21D 448 ± 20 15.7 ± 2.3 Serine/threonine protein phosphatases Cg17746 258 ± 32 4.3 ± 0.6 PPA-2C like puckered 228 ± 11 2.5 ± 0.2 Multifunctional twins 209 ± 11 2.0 ± 0.1 PPA-2A like Pp2A-29B 738 ± 28 1.9 ± 0.2 PPA-2A like Microtubule star 997 ± 46 1.3 ± 0.1 PPA-2A like Pp1-87B 318 ± 17 1.1 ± 0 PPA-1 like Pp1alpha-96A 332 ± 8 1.1 ± 0 PPA-1 like Accessory proteins, associated with anchoring, cellular localization or modulation of signaling Akap550 136 ± 6 2 ± 0.3 A-kinase anchor protein AKAP200 414 ± 30 0.35 ± 0.02 14-3-3-zeta 1,789 ± 42 2.6 ± 0.2 Diacylglycerol-activated PKC inhibitor CG32812 42 ± 4 2.7 ± 0.4 Calcineurin Rack1 6,463 ± 105 1.3 ± 0 Receptor for activated C-kinase ::: ::: {#T12 .table-wrap} Table 12 ::: {.caption} ###### Transcription factors and DNA-binding proteins that are abundant or enriched in tubule ::: Gene Signal Enrichment ------------ ------------ ------------- CG10278 175 ± 7 24.1 ± 11.5 CG5093 50 ± 4 19.3 ± 6.2 pnt 63 ± 5 17.5 ± 4.8 CG2779 5771 ± 317 16.8 ± 0.7 Ptx1 183 ± 8 12.7 ± 2.2 Ets21C 51 ± 17 9.8 ± 3.1 CG4548 91 ± 4 8.8 ± 4.3 HLH4C 6 ± 1 7.7 ± 6.9 fkh 266 ± 26 7.2 ± 1.1 hth 162 ± 13 7.2 ± 0.7 CG4566 17 ± 2 7.1 ± 4.2 bowl 71 ± 5 7.1 ± 0.7 CG4037 5 ± 1 6.7 ± 2.4 tap 5 ± 1 6.0 ± 3.0 CG6913 5 ± 2 6.0 ± 5.5 CG3950 287 ± 21 5.4 ± 0.9 Awh 21 ± 4 4.8 ± 1.4 CG1162 8 ± 1 4.7 ± 2.1 ct 145 ± 12 4.6 ± 0.8 CG14202 10 ± 1 4.6 ± 1.5 tsh (ae) 65 ± 5 4.6 ± 0.8 CG9952 45 ± 11 4.5 ± 0.6 sv 16 ± 2 4.3 ± 1.8 fd59A 11 ± 3 4.3 ± 1.7 CG11914 31 ± 4 4.2 ± 1.7 slp2 4 ± 2 4.1 ± 3.1 Lim3 13 ± 3 4.0 ± 1.1 CG6419 18 ± 3 4.0 ± 0.4 Tis11 337 ± 17 3.9 ± 0.6 nvy 27 ± 4 3.9 ± 1.1 ::: ::: {#T13 .table-wrap} Table 13 ::: {.caption} ###### Genes with known significance to tubule function, but very low abundance/enrichment scores ::: Gene name Signal Enrichment ----------- -------- ------------- NOS 1 ± 0 0.2 ± 0.04 trpl 9 ± 2 0.03 ± 0.01 trp 3 ± 1 0.02 ± 0.01 ::: ::: {#T14 .table-wrap} Table 14 ::: {.caption} ###### Drosophilatubule as a model for human genetic disease ::: Gene Affymetrix signal Enrichment Blast probability OMIM reference Human disease Available fly stocks --------------- ------------------- ------------ ------------------- ---------------- --------------------------------------------------------------------- -------------------------------------------------------- CG10226 290 ± 25 28.3 1.00E-184 171050 Colchicine resistance CG7402 99 ± 4 26.9 2.00E-40 253000 Mucopolysaccharidosis IVA Ir 1,302 ± 112 14.2 1.00E-76 600359 Bartter syndrome, antenatal, 601678 ry 655 ± 44 13.0 1.00E-184 607633 Xanthinuria, type I, 278300 Ptx1 183 ± 8 12.7 6.00E-38 602669 Anterior segment mesenchymal dysgenesis and cataract, 107250 Fmo-1 131 ± 11 12.0 9.00E-27 136132 \[Fish-odor syndrome\], 602079 CG4484 504 ± 50 12.0 1.00E-49 606202 Oculocutaneous albinism, type IV, 606574 DS00004.14 759 ± 54 10.6 1.00E-123 603470 Citrullinemia, 215700 CG9455 355 ± 40 9.0 1.00E-42 107400 Emphysema; emphysema-cirrhosis, hemorrhagic diathesis due to CG5582 825 ± 49 8.5 1.00E-69 607042 Ceroid-lipofuscinosis, neuronal-3, juvenile, 204200 Cyp4d2 1,008 ± 70 8.3 1.00E-27 107910 Gynecomastia, familial, due to increased aromatase activity CG7433 1,364 ± 50 7.4 1.00E-153 137150 GABA-transaminase deficiency CG1140 894 ± 26 7.3 1.00E-176 245050 Ketoacidosis due to SCOT deficiency CG9547 860 ± 34 7.0 1.00E-164 231670 Glutaricaciduria, type I PhKgamma 2,665 ± 152 6.9 1.00E-111 172471 Glycogenosis, hepatic, autosomal CG4623 382 ± 37 6.8 4.00E-28 606598 Charcot-Marie-Tooth disease, mixed axonal and demyelinating l(3)j7B3 CG12370 203 ± 17 6.7 5.00E-40 138033 CG15556 221 ± 12 6.4 6.00E-12 602851 Convulsions, familial febrile, 4, 604352 KCNQ 101 ± 0 6.4 1.00E-108 602235 Epilepsy, benign, neonatal, type 1, 121200; myokymia with neonatal CG17119 852 ± 28 5.7 6.00E-74 606272 Cystinosis, atypical nephropathic; cystinosis, late-onset juvenile CG7408 168 ± 6 5.6 3.00E-27 300180 Chondrodysplasia punctata, X-linked recessive, 302950 Spat 724 ± 39 5.1 2.00E-88 604285 Hyperoxaluria, primary, type 1, 259900 EP(x)1365 CG8743 1,001 ± 44 4.9 1.00E-100 605248 Mucolipidosis IV, 252650 CG14593 59 ± 13 4.9 2.00E-33 131244 ABCD syndrome, 600501; Hirschsprung disease-2, 600155 CG1673 911 ± 142 4.8 1.00E-100 113530 Hypervalinemia or hyperleucine-isoleucinemia (?) Ndae1 531 ± 34 4.7 1.00E-184 603345 Renal tubular acidosis, proximal, with ocular abnormalities, 604278 CG7834 3441 ± 106 4.3 8.00E-80 130410 Glutaricaciduria, type IIB, 231680 EP(2)2553, l(2)k00405 Pvr 294 ± 26 4.2 6.00E-69 164770 Myeloid malignancy, predisposition to CG12030 887 ± 51 4.1 1.00E-124 606953 Galactose epimerase deficiency, 230350 Mdr49 239 ± 25 4.0 1.00E-184 171060 Cholestasis, familial intrahepatic, of pregnancy, 147480 l(2)k05224 CG4685 563 ± 19 4.0 1.00E-129 271980 Succinic semialdehyde dehydrogenase deficiency EP(2)2545, l(2)k08713 CG12338 774 ± 16 3.9 4.00E-40 124050 CG12582 183 ± 14 3.8 1.00E-142 248510 Mannosidosis, beta- l(2)k10108 Reg-3 463 ± 24 3.8 1.00E-184 274270 Thymine-uraciluria Cyp12c1 73 ± 5 3.8 2.00E-34 124080 Aldosterone to renin ratio raised; hypoaldosteronism, congenital, Fur1 724 ± 29 3.7 1.00E-163 162150 Obesity with impaired prohormone processing, 600955 Cyp9c1 258 ± 14 3.7 7.00E-53 274180 Thromboxane synthase deficiency l(3)05545 Drip 318 ± 16 3.6 1.00E-37 154050 Cataract, polymorphic and lamellar, 604219 CG8654 274 ± 29 3.6 2.00E-62 607096 Hypouricemia, renal, 220150 Cyp9f2 1,700 ± 60 3.6 1.00E-69 124010 CYP3A4 promoter polymorphism; CYP3A4-V ERR 239 ± 21 3.5 5.00E-29 313700 Androgen insensitivity, 300068; breast cancer, male EP(3)3340 CG3603 94 ± 7 3.4 5.00E-20 222745 DECR deficiency (2) (?) CG9232 877 ± 20 3.4 1.00E-118 606999 Galactosemia, 230400 CG8417 502 ± 31 3.2 3.00E-71 154550 Carbohydrate-deficient glycoprotein syndrome, type Ib, 602579 EP(2)0844, EP(2)2192, EP(2)2358, l(2)05428, l(2)k06503 CG4663 439 ± 14 3.2 2.00E-29 601789 Adrenoleukodystrophy, neonatal, 202370; Zellweger syndrome, 214100 Cat 4,316 ± 88 3.2 1.00E-184 115500 Acatalasemia Prominin-like 308 ± 24 3.0 1.00E-20 604365 Retinal degeneration, autosomal recessive, prominin-related EP(2)0740 Genes that are abundant (Affymetrix signal \> 50) and enriched (\> 3 times) in tubule, and which are also closely similar (Blast probablility \< 10^-20^) to genes mutated in human genetic diseases, as described in the Homophila database \[99\]. OMIM reference refers to entries in the Online Mendelian Inheritance in Man database \[100\]. :::	PubMed Central	2024-06-05T03:55:48.524056	2004-8-26	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522876/", "journal": "Genome Biol. 2004 Aug 26; 5(9):R69", "authors": [ { "first": "Jing", "last": "Wang" }, { "first": "Laura", "last": "Kean" }, { "first": "Jingli", "last": "Yang" }, { "first": "Adrian K", "last": "Allan" }, { "first": "Shireen A", "last": "Davies" }, { "first": "Pawel", "last": "Herzyk" }, { "first": "Julian AT", "last": "Dow" } ] }
PMC522877	Background ========== Microarray experiments are often used to identify potentially relevant genes in biological processes. By determining which genes are differentially expressed between different states, for example, hypotheses can be developed as to the role of those genes in the underlying biological mechanism \[[@B1]-[@B4]\]. However, the fact that microarrays simultaneously assess the expression of tens of thousands of genes makes it difficult to extract pertinent information from background noise. With a multitude of variables, it is easy to generate a high percentage of false positives, and validation is expensive and time-consuming. This issue is aggravated by the high cost of microarrays and often by the difficulty of obtaining enough biological or clinical samples, causing microarray experiments to be performed on a smaller scale than desirable in almost all cases. For exploratory analysis in particular, very few biological or technical replicates are run at present. For a two-class comparison, three-by-three or smaller experiments are not uncommon. For brevity, we will use the notation \'NvN\' or \'N by N\' to denote a two-group comparison with N arrays vs N arrays. Overall, the need for a large sample size is acute for expression profiling studies. The number of arrays needed in a study depends on many factors, including the study design, the magnitude of biological variation in the samples, technical variability introduced in the experiment, and the desired level of sensitivity and specificity for differential expression. Several studies have examined this issue. A model with additive and multiplicative noise was used to derive the number of samples necessary for detecting fold changes of given magnitude when false-positive and false-negative rates are specified \[[@B5]\]. The difficulty, however, is that parameters describing technical and biological variations must be estimated for the model, which is not an easy task. When 16 public datasets, mostly from cancer studies, were examined using a repeated sampling approach \[[@B6]\], it was observed that stable results for differentially expressed genes are not obtained until at least five biological replicates are used and that 10-15 replicates are needed for sufficient stability. This is consistent with the results obtained in \[[@B7]\]. According to these criteria at least, many microarray studies are vastly underpowered. From the perspective of analysis, it is always desirable to have sufficient data. Some data analysts may even insist on a minimum number of samples before starting statistical analysis. However, when practical considerations limit the sample size, it is important to work with the given data in an optimal manner to extract as much information as possible. In the context of finding differentially expressed genes, the null hypothesis for each gene is that it is not differentially expressed between two groups, usually against the two-sided alternative hypothesis that the gene is up- or downregulated. The most commonly used statistical test in this setting has been the two-sample t-test, although other similar statistics such as the signal-to-noise ratios \[[@B1]\] have often been used as described below. There are a variety of statistical issues involved with identifying differential expressed genes, such as the adjustment of p-values for multiple testing \[[@B8]\] and the use of the false-discovery rate \[[@B9]\]. Ideally, the joint distribution of the test statistics should be considered, in order to account for correlation among the genes \[[@B10]\], but in practice, because of the difficulties associated with the number of genes being many times that of the samples, most testing procedures are carried out in a univariate manner for each gene \[[@B11]\]. The method we introduce here also performs a test independently for each gene and ignores correlation among genes. A fundamental difficulty in drawing reliable conclusions from a small number of samples lies in accurate estimation of the gene-specific variances, or the variance of a difference in mean expression levels per gene, with which to determine the statistical significance of observed changes in expression. Because variances based on a very small number of samples tend to fluctuate wildly as a result of randomness in sampling from a population, our ability to assess differential expression is drastically impacted. A naive application of standard methods used for larger sample sizes can result in a large number of false positives for differential expression. For example, with a small sample size, the list of significant genes identified by the t-test or variations thereof is crowded by a large fraction of genes for which large t-statistics are due to underestimation of variance by chance. Many methods have been devised to address this problem. A popular approach has been some type of regularization of the t-test. A Bayesian framework for combining the variance estimate with a background variance associated with neighboring genes was developed in \[[@B12]\]; a method of pooling errors among genes in which expression values are similar is presented in \[[@B13]\]. In the popular significance analysis of microarrays (SAM) method, a small constant is added to the variance estimate to prevent it from getting too small \[[@B14]\]; empirical Bayes methods compensate for the lack of enough replicates by combining information across the arrays \[[@B15]-[@B17]\]. Nonparametric methods \[[@B18]\], analysis of variance approach \[[@B19]\], and Bayesian hierarchical models \[[@B20],[@B21]\] are also available. Some of these methods are compared in \[[@B22]\]. Whereas all the available methods attempt to improve the identification of differentially expressed genes essentially by gathering information across similar genes, we suggest another solution. We propose estimating the natural variance of individual genes using a large number of experiments performed previously. This provides a different and potentially more stable and accurate estimate of the variance for each gene than by simply looking at the variance of a small number of expression levels, especially in studies with very small sample sizes. Using these variances as the basis for determining differential expression offers an alternative method that can reduce the false-positive rate significantly. As the most effective method, we propose a hybrid method in which we combine the variance estimate from the current dataset with the estimate from previous experiments. This approach can also be incorporated in other settings, especially in a Bayesian framework with prior distribution for variance derived from the database. It can be applied more generally to other testing procedures such as ANOVA that benefit from more accurate estimation of gene-specific variances, and can be easily extended to the estimation of the covariance matrix in multivariate analysis. More reliable calculations of such variances based on many chips is becoming increasingly possible through large public databases of previous experiments. Public databases such as the Gene Expression Omnibus (GEO) \[[@B23]\] contain data from many chips, with the goal of gaining information from pooling data. GEO currently has thousands of chips, with a heavy skewing towards Affymetrix MG-U74Av2 and HG-U95 chips. Specifically, there are about a thousand HG-U95A chips and another thousand MG-U74Av2 chips, and these numbers are growing steadily (our gene-specific variances were calculated when the database held only 865 chips). Other large public databases include ArrayExpress \[[@B24]\], Yale Microarray Database \[[@B25]\], and Stanford Microarray Database \[[@B26]\]. GEO was selected as our primary source of reference because it had the largest compilation of single-channel microarray chips. We chose to analyze Affymetrix chips because the standardization of single-channel chips allows for easier cross-experiment comparison than dual-channel chips. The dual-channel chips are often custom-made and lack consistency in the genes represented; more important, different experiments use different reference channels, which makes it difficult to compare across experiments. Results ======= Comparing various methods ------------------------- We compared the performance of four methods in accurately assessing differential expression of genes: the standard t-test, the new GEO-adjusted method, the regularized t-test, and a hybrid method combining the GEO method and the regularized t-test. The primary difference between these methods lies in the denominator of each method\'s t-statistic. The GEO-adjusted method replaces the sample variance estimate in the denominator with the gene-specific variance calculated from the GEO database (details for the calculation of the variance, which can be either global or pooled, are described in Materials and methods). Hence, the genes are sorted using the modified t-statistic: ![](gb-2004-5-9-r70-i1.gif) where μ~1i~, μ~2i~are the means for the groups 1 and 2, respectively, for the ith gene, n~1~and n~2~are the sample sizes in the groups 1 and 2, and σ^2^~GEO,1~is the gene-specific variance from the GEO database. The regularized method we used added a small constant to the denominator of the t-test, ranking genes based on the modified t-statistic: ![](gb-2004-5-9-r70-i2.gif) where σ~0~is the fifth percentile of all variances (σ~0~can also be calculated to minimize the coefficient of variation of the statistic \[[@B14]\]). The regularized t-test smoothes out the effects of underestimated variances and therefore returns a more reliable assessment of differentially expressed genes in small samples than the standard t-test. Finally, we introduce a hybrid algorithm that combines the GEO method with others through a voting mechanism. This provides a portable solution that can be combined with a variety of other testing procedures and could potentially improve the performance of any other algorithms designed to determine differential expression in experiments with small sample sizes. Testing procedure and dataset ----------------------------- To compare the effectiveness of the methods, we determined lists of differentially expressed genes in order of significance by applying each procedure to a large number of subsets of arrays of a given size. These genes were then compared with the \'master\' list of differentially expressed genes to assess the accuracy of the method. Because we generally do not know the correct ordering of genes with differential expression, we substituted the list obtained by the t-test analysis for the full dataset as the master list; with a sufficiently large dataset, this master list is a close approximation to the true list. Thus, we used the large dataset to compute a true t-statistic for each gene and then treated random small subsamples of the arrays from the dataset as simulated observed datasets from which we could compute estimated ranks for small subsamples. Because exploring all realizations of possible subsets for a large full dataset would be prohibitively time-consuming (for example, more than 10^8^combinations of 3v3 subsets for our first dataset), we sampled repeatedly for subsets until we obtained convergent results. We then compared these lists of differentially expressed genes with the master list for overlaps or correlations in the orderings. Once the size of the subsets approaches the size of the full dataset, there can be a substantial overestimation in the overlap of genes, owing to the fact that the master list is generated using the dataset from which the subsamples are derived. However, this effect appears to be negligible in our simulations because of the large size of our full dataset and the small subsample sizes that are of our interest. The dataset primarily used for testing was a prostate cancer dataset \[[@B27]\] that had 50 normal samples and 52 tumor samples, with follow-up tests performed on a smaller Duchenne muscular dystrophy dataset \[[@B28]\] to confirm results. In our subsampling process, a small number of patients are randomly selected from each group and a variety of methods were used to determine a list of differentially expressed genes. We are mainly interested in very small sizes of one to three samples per group. For concreteness, we focus on the results for 2v2 comparisons first, but we also describe 1v1 and 3v3 comparisons. Note that large datasets are utilized here solely for the purpose of evaluating the method and that the method is designed to be used for studies with small samples. Numerical results with a GEO-adjusted t-test ---------------------------------------------- The first measure that was used to assess the effectiveness of the GEO-adjusted method was the correlation between the rank of the top genes returned by various methods and the true rank of those genes. This method was also used in \[[@B6]\]. In this measure, the standard t-test was compared to the GEO-adjusted method. The behavior of the correlation coefficient was tracked as the number of genes being analyzed was increased and the averaged values over many simulations are shown in Figure [1](#F1){ref-type="fig"}. If the method were perfectly effective and ranked genes in the same order as their true ranking according to the master list, the correlation should be 1. However, the correlation coefficients were surprisingly low. This reflects the great difficulty of obtaining accurate or stable lists of differentially expressed genes from small sample sizes. Nonetheless, Figure [1](#F1){ref-type="fig"} reveals that the correlation improves for the t-test as the sample size increases, and that the results of GEO tend to correlate better with the master list than the results of 2v2, 3v3, 4v4, or 5v5 t-test. To further assess the reliability of the results, tests were conducted to determine the number of top 50 genes from the master list that were accurately returned using various methods. In Figure [2](#F2){ref-type="fig"}, this is plotted as a function of the list length generated by these methods, at 10, 50, 100, 150, 200, 250 and 300. For example, a list of 100 genes from the 2v2 GEO method contains just over 10 genes from the top 50 genes from the master list. Again, the low overlap clearly illustrates the difficulty of obtaining an accurate list of significant genes. We believe the low numbers to be partly due to the nature of heterogeneous samples in our test datasets (see Discussion); therefore, we focus more on the trend among the various methods here. This metric indicated that the GEO method is considerably more reliable than the t-test at determining differentially expressed genes in small sample sizes. Compared to a simple t-test, the GEO method performs substantially better, returning results from a 2v2 test that are comparable to the results returned by a 5v5 experiment using t-test. Using GEO variances on a 2v2 test returns 231% more of the top 50 genes than the unadjusted t-test. While we are not suggesting that a simple t-test is a recommended method of assessing differential expression in such small sample sizes, it illustrates the potential of this method. Using gene-specific variances developed from GEO databases is clearly more accurate than the variances that an uninformed t-test derives from small samples. We do not plot the error bars for each measurement in the figures owing to space constraints, but we have verified in the important cases that the separation between the curves is significant. The GEO-adjusted method also compared favorably to a regularized t-test. By smoothing out the variance estimates, the regularized t-test returns a more accurate assessment of differentially expressed genes than the standard t-test. Thus the gains from the GEO method over the regularized t-test were less substantial than over the standard t-test but still significant, especially for shorter gene lists. Improving our ability to reliably detect the differentially expressed genes with a short list is generally more valuable than doing so with a longer list simply because these genes at the top are the ones that an investigator examines most closely. As shown in Figure [3](#F3){ref-type="fig"}, the average gain of the 2v2 GEO sample versus the 2v2 regularized t-test in those three areas (50, 100 and 150 genes) was more than 30%. The performance of GEO on a 2v2 analysis seems roughly comparable to the performance of the regularized t-test on a 3v3 sample analysis. Superior performance of a hybrid method --------------------------------------- One of the greatest advantages of the GEO method is that it can be combined with other methods. Because the regularized t-test and the GEO method both use different, yet effective, techniques to smooth out variance, they can both contribute to the differential expression analysis. By using a voting method that weights and averages the results returned from the regularized t-test and the GEO method, the performance improves further (see Materials and methods for details). The results of a 2v2 chip analysis using our voting method nearly match the performance of a 4v4 regularized t-test analysis, which is quite promising (Figure [3](#F3){ref-type="fig"}). As before, our incidence of the top 50 genes in the top 10 listed, top 50 listed and top 100 listed are improved. The voting method returns 88% more of the top 50 genes than the regularized t-test alone. We also see the greatest improvement in the larger sets of genes, thus negating one of the weaknesses of the GEO-adjusted method. By combining the advantages of the regularized t-test and the additional information from the gene-specific variances, we are able effectively to pare the required number of chips in this case and to elicit better results from the chips we do have. Further details are provided in the Materials and methods section. Tests were also performed on other sample sizes, namely 1v1 and 3v3. Although we view the first case especially as an inadequate design and do not recommend it, we have found that investigators are sometimes forced to perform analysis on such a small number of samples. We are therefore interested in improving the effectiveness of such exploratory analysis, the results of which must be verified using other techniques such as quantitative reverse transcription PCR (QRT-PCR). In our 1v1 analysis, we compared the GEO method to three methods of ordering genes on the basis of differential expression: fold ratio, y/x; percent changes relative to the mean expression levels, (x - y)/((x+ y)/2); and z-score based on local variance correction (using locally weighted polynomial regression) across signal intensity, as implemented in SNOMAD \[[@B29]\]. Basic filtering of low expression was performed at the beginning. In the example dataset, the z-scores give slightly better results than the percent changes, which in turn were better than simple fold ratios. However, as shown in Figure [4](#F4){ref-type="fig"}, the GEO method returns 60% more of the top 50 genes than the best of the first two standard methods and generally returns superior results, almost on the same scale as a 3v3 regularized t-test. The method based on the z-scores performs slightly better than either of the standard methods, but GEO still returns 57% more of the top 50 genes. In the 1v1 case, the voting method proves useful, improving the results of both methods. By combining the z-score method and GEO\'s rankings, the results are superior to a 3v3 regularized t-test analysis. The voting method captures 83% more of the top 50 genes than the best of the standard methods. These results reflect the success of the voting method in combining GEO\'s rankings with a variety of other methods to significantly improve the overall performance. The results from the regularized t-test and GEO method were also compared on 3v3 comparisons. Whereas GEO still returns more reliable estimates than the regularized t-test, the improvement is smaller than in the case of the smaller sample size comparisons. In the 3v3 case, the GEO-method results are comparable to those of a 4v4 regularized t-test, returning 17% more of the top 50 genes than the 3v3 regularized t-test. However, we do find that the voting method again improves the results, returning very similar numbers of correct genes as a 5v5 regularized t-test (Figure [5](#F5){ref-type="fig"}). The voting method returns 41% more of the top 50 genes than the 3v3 regularized t-test. The performance of the GEO method does not seem to be influenced greatly by the number of samples in each group. This is because the gene-specific variance estimates are fixed and adding additional samples only impacts the mean estimates for each group. In contrast, in the regularized t-test method, adding additional samples to each group refines the estimates of both the means and the variances of each group. This factor is the fundamental reason that the regularized t-test improves quickly as the number of samples is increased whereas the GEO method does not. However, GEO performs strongly even with only one sample in each population and generates results that are comparable to a 3v3 regularized t-test analysis. This indicates that the greater weakness in the small-sample t-test lies in inaccurate variance estimates, and that stable, accurate estimates of gene-specific variance can greatly improve analysis. These results are summarized in Figure [6](#F6){ref-type="fig"}, which compares the performance of the standard t-test, the regularized t-test, the GEO method, and the voting method across sample sizes. The voting method is substantially better in all cases. Looking at the Duchenne muscular dystrophy dataset also provides us with corroboration of the usefulness of this method. In this situation, the dataset is much smaller (11 normals vs 12 DMD). Because two samples capture a much higher percentage of the data in 11 chips than in 50 chips, we expect the usual tests on subsamples to naturally provide results more similar to the master list. Therefore, we expect to see less of an improvement from GEO than in our cancer dataset. As before, we see the GEO results consistently providing better results in the smaller sets of genes than the standard t-test, returning 33% more of the top 50 genes in the 2v2 case (Figure [7](#F7){ref-type="fig"}) and 40-170% more of the top 10, 50, 100 and 150 genes in the 1v1 case (Figure [8](#F8){ref-type="fig"}). While the regularized t-test seems to outperform the GEO method, combining the results of both using our voting method again returns superior results. For example, averaging the ranks in the 2v2 case returns us 134% more of the top 50 genes than the regularized t-test alone and 240% more than the standard t-test (Figure [7](#F7){ref-type="fig"}). In the 1v1 case, the voting method clearly outperforms either the GEO method or the local z-score method (as implemented in SNOMAD) alone, providing results that seem roughly similar to those returned by a 2v2 regularized t-test analysis. These results indicate that, as shown in the cancer dataset, improved results can definitely be attained through incorporating gene-specific variance in differential expression analysis. Most important, because the GEO method can be combined with regularization methods through a voting procedure, it can be used to improve results regardless of how it individually performs on a dataset. Discussion ========== Number of chips --------------- For this method to be successful, a significant number of previously run chips must be available. As public databases grow in size and number, this limitation will gradually diminish, but not all chip types currently have enough available chips to use this method. Whereas the most popular chip types (such as Affymetrix HG-U95A) have hundreds of previously run chips available, it is more difficult to find databases of the less popular ones. In an attempt to test for the number of chips sufficient to utilize this method, variance analysis was performed. In Figure [9](#F9){ref-type="fig"}, we plot the variance estimate as the number of chips used in the estimation increases, for one realization of the chip ordering. Because genes at different intensity levels may behave differently, we sorted the genes by their expression levels and selected four genes, one from the middle of each quartile. As seen in each case, the variance calculated from many chips tends to converge as the number of chips grows. Generally, the variances seemed to settle near their final values once 250-300 chips are gathered. After averaging over a large number of realizations in the chip order, we find that the variance settles near its final value at 250 chips, deviating less than 5% in either direction as more chips are gathered. While it is difficult at this time to find 300 chips of similar type and tissue, it should become easier to find datasets that are more specifically correlated with the experimental set as more data are accumulated in public databases. This would allow for more useful baselines to be established in calculating gene-specific variance, and would probably substantially improve the results. Comparing across multiple tissue types -------------------------------------- When trying to estimate the gene-specific variances for a particular experiment, the best approximation would come from a database of similar experiments. Because gene expression profiles have the potential to vary widely in cell and tissue type, examining many other chips of the same tissue type should provide the best indication of the baseline variance. For example, it would make most sense to draw on a large database of cancer chips to derive relevant information for a cancer dataset. Unfortunately, because of the dearth of large datasets that match each other in tissue type, chip type, and post-processing, it is difficult at this time to test this theory. Because our public databases do not yet contain enough chips sorted by tissue type to perform this procedure, we are forced to mix all the chips of any given type together. Yet, even with only a database of totally unrelated chips, we still saw a significant increase in performance, even over already improved methods such as the regularized t-test. If our gene-specific variances were based on even more reliable estimates (such as samples from the same tissue or same disease), the performance of the GEO method would probably be increased. As public databases grow in size and organization, this should become increasingly possible. Comparing across multiple chip types ------------------------------------ We have shown here how to derive more stable variances based on chips of the same type. The problem is much more complicated when multiple chip types are involved. In fact, we have observed that even different generations within the same platform do not give concordant results. When the same tissue samples were hybridized on both HG-U95A and HG-U133A chips, the dominant feature in the data was the chip type rather than the sample characteristic, and the lists of differentially expressed genes differed substantially between the two cases. Standardizing across these two types can be done for a portion of the genes but it is an involved process (Hwang KB, Kong SW, Greenberg SA, P.J.P., unpublished work). Efficiently combining data from single-channel and double-channel arrays is even more difficult. A more comprehensive database with a larger number of arrays spanning a greater variety of experiments would alleviate the problem to some extent, but methodologies for integrating data from multiple platforms will be essential, not only for better estimation procedures for differential expression but for other purposes as well. Need for standardization in public databases -------------------------------------------- Public databases are an important resource for investigators to consider. With the stores of chips accessible online, valuable information concerning genes can be compiled and used to supplement new studies and avoid duplication of effort. This methodology would be improved by modification to the databases. Mainly, it is very important to start gathering the raw files instead of processed files. With Affymetrix chips, for example, .cel files should be stored, so that they can be processed using the latest methodology and maintain their usefulness. Already, many of the chips in the GEO database are less useful because they only report values processed through MAS 4.0, an outdated methodology, and comparing these with values generated through MAS 5.0 introduces another source of variation. In addition, the chips should be categorized and sorted according to tissue type, to further facilitate grouping and analysis. These modifications would improve the ability to use previously run chips, thus countering the high costs associated with microarray experiments and enabling the sharing of information to accelerate progress. Thinking about how to take advantage of these databases could provide further improvements to methodologies and enable more tools to be used to study gene functions. Conclusions =========== This work proposes that value lies in pooling information from previous studies. Specifically, gene-specific information can be collected from public databases housing many chips, supplementing new studies and ensuring more reliable results. We show that compiling information from databases provides us with a different and potentially more accurate estimate of gene-specific variance, improving our differential expression analysis in small samples. In addition, because this improvement seems largely independent of the method of analysis, we are able to combine it with regularization in a voting method, leading to superior results. There were particularly strong improvements in the identification of the smallest groups of most differentially expressed genes, which would probably be deemed most important by an investigator as they are easiest to validate. Overall, the scale of the improvement is significant, as it allows investigators to halve their costs in some cases and still retain similar accuracy. The same approach can also be formulated in other settings, especially in the Bayesian framework in which priors for the gene variances may be estimated from previous datasets. Furthermore, as public databases are steadily growing in size, we expect refinement of this method to deliver greater success in the future. Regardless of what method an investigator might use, public databases are clearly a useful source of information and should prove useful in supplementing microarray studies. Materials and methods ===================== Because the very nature of public datasets implies that many of the chips have been generated and processed in different manners, standardization of the data is paramount. To maintain comparability, the chips were filtered to remove any chips processed with an algorithm other than MAS 5.0. After removing other unusable chips (such as duplicates and abnormally processed chips) 471 HG-U95A chips remained. Normalization of all of these chips is crucial, in order to guarantee that scales are similar. In an effort to preserve the general characteristics of each chip, conforming their scales while allowing for some chip-by-chip variability, experiments with multiple methods of normalization were carried out. The two major types included normalizing the trimmed mean and trimmed variance of each chip and using percentage ranks instead of numerical expression levels. In the first case, all of the data points were adjusted to align the mean and variance of the middle 90% of values. In the percentage ranks method, the values were assigned percentiles, removing most normalization effects. In addition, a scale was generated that related the percentile with the average rank change of that percentile. The average rank change for a gene in the middle of the scale was significantly larger than the average rank change at either extreme. This scale was used to adjust the variances on the basis of the rank. Because the results from both normalization methods were fairly similar, only the results of the trimmed mean, trimmed variance experiments are reported here. After all of the chips were normalized, the gene-specific variance was calculated. These variances were calculated in two separate ways, using a global variance and a pooled variance: ![](gb-2004-5-9-r70-i3.gif) ![](gb-2004-5-9-r70-i4.gif) where, for each gene, x~ij~is the expression level of array iin experiment set j; ![](gb-2004-5-9-r70-i5.gif) the mean in the experimental set j; ![](gb-2004-5-9-r70-i6.gif) is the mean in all arrays; Dand D~j~contain the indices for the experimental sets and the arrays in the jth set, respectively. The global variance tends to reflect the degree that a gene may vary between different tissue types and diseases while the pooled variance reflects the degree that a gene tends to vary within each experiment. The global variance proved slightly more effective in the cancer dataset, while the pooled variance was more effective in the muscular dystrophy dataset. This seemed to be correlated to the composition of our GEO background datasets. Our set of 471 GEO chips contained 210 cancer chips but only 42 muscle chips. Because a large proportion of the total chips were cancer chips, a global variance may have more accurately represented the information in the whole dataset. However, because so many of the GEO chips were non-muscle, readjusting them into a pooled variance format may have provided a better gene-specific assessment of general expression. We further filtered the variance calculations to eliminate artifacts created by improperly processed chips along with biases from experimentation (that is, if a certain experiment produced uniformly high values for a specific gene). Thus, the highest and lowest 10% of values for each gene across the full set of GEO arrays was trimmed off for the variance calculation. The 10% parameter was chosen experimentally, by tracking how stable variance calculations were as various percentages were trimmed. The statistical properties of these variance estimators are difficult to show rigorously. If the samples from the GEO datasets can be assumed to come from the same population as those in the current study, the estimators should be unbiased and the proposed test statistic should behave as N(0,1) asymptotically. Because the GEO data are an aggregate of many experiments under different conditions often processed differently, we cannot assume the same underlying distribution in general and hence we do not know if the estimators necessarily approach the true variance. However, these estimators appear to be reasonably good approximations to the \'true\' variance as demonstrated by the numerical results, and they certainly perform better than estimates based only on the current data. A master list of the most differentially expressed genes in the dataset was determined by t-test analysis. Then, the various methods were compared with each other through a process of subsampling to determine how accurately the results reflect the master list. After two samples from each group were randomly selected, all the genes were filtered out that did not have an expression level above 100 in any of the samples. The goal was to lower false positives among the non-GEO methods, as their results could easily be influenced by small expression levels that by chance ended up with virtually no variance and thus were assigned large t-statistics. After processing in this way, the top genes derived using the t-test, the regularized t-test, and the GEO-adjusted method were compared with the master list to determine their effectiveness. This subsampling procedure was repeated 500 times for each experiment, and the results were averaged. These methods are outlined in the Results section. Although the GEO-adjusted method was superior to both the t-test and the regularized t-test, the greatest success was found by averaging the results of the regularized t-test method and the GEO method. By averaging the ranks that we receive using GEO and using the regularized t-test set, our results are improved concerning our most important genes. This is not seen when the results of the simple t-test and GEO are combined, because the lists produced by the simple t-test are simply too inaccurate. However, as GEO and regularized t-test produce lists that are similar in quality, yet different in nature, a boost can be obtained by averaging the lists. Because using just the GEO variances ignores some of our experimental data and using just our experimental variances ignores global data, it seems that an averaging or voting procedure is a superior way to optimize results. In particular, the system that we used averaged 75% of the value of the lower rank (nearer the top of the list) with 25% of the value of the higher rank. A final score was obtained by combining the results from each method in this way, and the genes were re-ranked on the basis of this score. By using the 75%/25% ratio, genes that have a particularly high ranking on one of the methods are given slightly more importance than genes that have average rankings in both methods. Empirical testing of a number of combinations showed that the 75%/25% combination returned superior results, although all combinations experimented with returned results that were better than either method alone. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The correlation between the rank of the top genes with their \'true\' rank, based on the \'master\' list from the full data. The x-axis is the length of the gene list being compared. The correlation of the GEO method is clearly superior to the correlation of the simple t-test. ::: ![](gb-2004-5-9-r70-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### A comparison of the reliability of differential expression results returned by simple t-test and the GEO method. The number of the top 50 differentially expressed genes from the master list that are found in the gene list of length 10, 50, 100, 150, 200, 250 and 300 is indicated on the y-axis. The GEO results based on a 2v2 sample are comparable to the results returned by a 5v5 sample t-test. ::: ![](gb-2004-5-9-r70-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### A comparison of the reliability of differential expression results returned by regularized t-test, the GEO method and the voting method in a 2v2 sample comparison. The number of the top 50 differentially expressed genes from the master list that are found in the gene list of length 10, 50, 100, 150, 200, 250 and 300 is indicated on the y-axis. The GEO results based on a 2v2 sample are clearly superior to the 2v2 regularized t-test results, and roughly comparable to the results of the 3v3. The voting method combining the results improves the results to a level almost comparable to a 4v4 regularized t-test. ::: ![](gb-2004-5-9-r70-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### A comparison of the reliability of differential expression results returned by the GEO method, a few standard methods, and the voting method in a 1v1 sample comparison. The number of the top 50 differentially expressed genes from the master list that are found in a gene list of length 10, 50, 100, 150, 200, 250 and 300 is indicated on the y-axis. The GEO results based on a 1v1 sample are clearly superior to the 1v1 results from the local z-score method (as implemented in SNOMAD), or from the percentage difference (using percent changes relative to the mean expression), and almost comparable to the results of the 3v3 regularized t-test. The voting method combining the results improves the results to a level superior to a 3v3 regularized t-test. ::: ![](gb-2004-5-9-r70-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### A comparison of the reliability of differential expression results returned by regularized t-test, the GEO method, and the voting method in a three sample by three sample comparison. The number of the top 50 differentially expressed genes from the master list that are found in the gene list of length 10, 50, 100, 150, 200, 250 and 300 is indicated on the y-axis. The GEO results based on a 3v3 sample are clearly superior to the 3v3 regularized t-test results, and roughly comparable to the results of the 4v4 regularized t-test (not shown). The voting method combining the results improves the results to a level almost comparable to a 5v5 regularized t-test. ::: ![](gb-2004-5-9-r70-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Summary of the performance of the four methods (standard t-test, regularized t-test, GEO method, and voting method). The bars indicate the percent improvement over the 2v2 standard t-test in identifying the top 50 differentially expressed genes. GEO performs better than the regularized t-test in smaller sample sizes, while the regularized t-test outperforms GEO in larger sample sizes. The voting method is substantially better in all cases. ::: ![](gb-2004-5-9-r70-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### A comparison of each method in 2v2 subsampling of a Duchenne muscular dystrophy dataset. The most positive results are clearly seen in the voting method combining the regularized t-test and GEO results. This method returns 134% more of the top 50 genes than the regularized t-test alone and 240% more than the standard t-test. ::: ![](gb-2004-5-9-r70-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### A comparison of each method in 1v1 subsampling of a Duchenne muscular dystrophy dataset. The most positive results are clearly seen in the two voting methods combining the GEO results with either the local z-score method (as implemented in SNOMAD) or the percentage difference method (using percent changes relative to the mean expression levels). These methods return 96% more of the top 50 genes than the standard method alone and 80% more of the top 50 genes within 100 genes. ::: ![](gb-2004-5-9-r70-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### The progression of the variance estimate as the number of chips used in the estimation increases, for one realization of the chip ordering. After all the genes were sorted by their intensity level, one gene was selected from the middle of each quartile. As seen in each case, the variance calculated from many chips tends to converge as the number of chips grows. Generally, the variances seem to settle near their final values once 250-300 chips are gathered. ::: ![](gb-2004-5-9-r70-9) :::	PubMed Central	2024-06-05T03:55:48.537136	2004-8-26	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522877/", "journal": "Genome Biol. 2004 Aug 26; 5(9):R70", "authors": [ { "first": "Richard D", "last": "Kim" }, { "first": "Peter J", "last": "Park" } ] }
PMC522878	Background ========== Structural genomics and structural proteomics involve the systematic structural analysis of large sets of proteins \[[@B1],[@B2]\]. Structural analysis requires protein samples of high quality and reasonable quantity \[[@B3]\]. Bacterial protein-expression systems, namely Escherichia coli, are well suited for preparing such samples at high throughput. Genetic manipulation of E. coliis easy and large amounts of recombinant protein can be expressed in a short time. However, low success rates have been reported for the expression of eukaryotic proteins in E. coli: only a small proportion of proteins can be successfully expressed, partly owing to the specific requirements of eukaryotic proteins in regard to the cellular environment \[[@B4]-[@B6]\]. Alternative expression systems such as yeast, insect cells/baculovirus or mammalian cell lines are being improved and have great potential to express larger sets of proteins in the amounts and purity required for structural analysis \[[@B7]\]. Cell-free expression systems represent another valuable alternative \[[@B8]\]. At the moment, these systems still require more experimental effort compared to expression in E. colicells. Consequently, one possible approach to structural proteomics for eukaryotic proteins is to study those that can be expressed in E. colifirst. A human cDNA expression library (hEx1) was constructed for parallel screening of protein function on high-density protein arrays \[[@B9],[@B10]\]. This library was cloned into a vector for expression of His-tag fusion proteins. The E. coliK-12 strain SCS1 was used for cloning the library and subsequent protein-expression experiments. A total of 193,536 clones were arrayed on protein-binding membranes and putative expression clones were detected immunologically, resulting in a smaller library of 37,830 putative expression clones \[[@B10]\]. This new library contains a large proportion of clones expressing His-tag fusion proteins from their cDNA inserts. Most of these expression products were found to remain in the insoluble fraction after cell lysis, which indicates that they form inclusion bodies. To identify clones that express their cDNA insert as a soluble, native folded protein, we established a high-throughput procedure for expression and purification of His-tag fusion proteins under non-denaturing conditions. This procedure was used to screen 10,825 clones of the hEx1 library for soluble expression products. Results ======= Clones expressing soluble protein --------------------------------- The hEx1 cDNA expression library was screened for expression clones on protein macroarrays. Using an anti-His-tag antibody, a subset of 37,830 clones was detected, as described before \[[@B10]\]. On the basis of a normalization experiment by oligonucleotide fingerprinting \[[@B11],[@B12]\], redundant clones were removed from this set of putative expression clones, and 10,825 clones were selected for further characterization (Figure [1](#F1){ref-type="fig"}). To identify soluble expression products, small-scale protein expression and purification experiments were performed in microplates in 1 ml cultures. Protein expression was routinely performed at 37°C for 7,316 clones. Because lower induction temperatures have been reported to increase the yield of soluble product for certain proteins \[[@B13]\], we carried out protein expression at 30°C and 37°C for a set of 284 clones. It was found that for some clones more soluble protein was obtained at 30°C, whereas for a smaller set of clones the yield was reduced. On the basis of these results we tested the remaining 3,509 clones at 30°C. Cells were lysed and aliquots were removed twice - before and after pelleting of cellular debris by centrifugation. These aliquots were termed \'whole\' and \'soluble\' protein extracts, respectively. Small-scale purification by metal chelate affinity chromatography was performed in batches of 96 in microplates, either manually or with the help of a pipetting robot \[[@B14]\]. Cellular protein extracts and purified protein samples were analyzed by SDS-PAGE (Figure [2](#F2){ref-type="fig"}). It was found that analysis of the purification eluates is more informative than analysis of the cellular protein extracts. Therefore, only the purification eluates were analyzed for most clones. For each sample, the size of the expression product, if any, and the yield of the recombinant product was recorded. The yield was roughly classified as follows: 0, no expression; 1, weak/doubtful expression; 2, moderate expression; and 3, strong expression. Only clones expressing soluble protein with a size of at least 15 kDa were selected. As found previously \[[@B10]\], the size of the expression product of a random cDNA expression clone is predictive of the reading frame of the cDNA insert. Most expression products with sizes of less than 15 kDa were found to be artificial products of cDNA inserts in the wrong reading frame, while expression products of at least 20 kDa were almost exclusively expressed from clones with cDNA inserted in the correct reading frame. Screening of the 10,825 hEx1 clones identified 1,866 clones (17%) expressing soluble protein of at least 15 kDa; 1,037 (10%) showed moderate or strong expression. Sequence analysis ----------------- Clones expressing soluble protein with a size of at least 15 kDa were subjected to DNA tag-sequencing, starting from the 5\' ends of the cDNA inserts. For 1,588 clones, sequences of at least 200 base-pairs (bp) of good quality were obtained. Of these sequences, 1,509 (95%) could be matched to transcript sequences from the Ensembl database \[[@B15]\], using the program cross\_match \[[@B16]\]. These transcripts correspond to 1,105 different genes. By matching their sequences to Ensembl, clones were assigned to human proteins and genes and clones containing complete open reading frames (ORFs) were identified. Transcript sequences from the Ensembl database are annotated with start and end positions of ORFs. Annotation of the ORF start position in Ensembl depends on experimental data from other databases and is not determined automatically. Many transcript sequences in the Ensembl database were generated automatically using cDNA sequences and exon-detection algorithms. If such a transcript is novel and does not correspond to known proteins, the ORF start position cannot be determined reliably by the automated annotation process of Ensembl. The annotation will often assign an ORF starting at position 1 to such transcripts; this is the case for 33% of transcript sequences in the Ensembl release 20.34c. To determine which cDNA clones contain complete ORFs (full-ORF clones), the Ensembl database was used, despite the limitation outlined above. Of 1,509 cDNA clones, 538 (36%) were identified as full-ORF clones, as their 5\'-tag sequences align to an Ensembl transcript sequence at a position upstream of the ORF start position on that sequence. These clones, representing 375 distinct transcripts, were annotated as containing a complete ORF, as the cDNA for the hEx1 library was constructed by oligo(dT) priming and is therefore assumed to contain the 3\' end of their transcript templates. For expression of cDNA inserts as His-tag fusion proteins, the respective cDNA insert has to be cloned in-frame to the vector-encoded start codon and His-tag. The reading frames of the clones\' cDNA inserts were determined from the positions of Ensembl transcript and vector sequences aligned to the clones\' sequences (see Materials and methods). We determined the reading frame of 1,447 of the 1,509 clones and found that 1,014 (70%) of the sequences were cloned in the correct frame with respect to the vector. Observed expression product sizes compared to prediction -------------------------------------------------------- The complete clone insert sequences are unknown as only partial 5\'-tag sequences were generated. However, if a clone is matched to an Ensembl transcript sequence, it is possible to construct a putative predicted insert sequence by combining the experimentally derived DNA sequence and the Ensembl transcript sequence. Such a strategy can lead to wrong results if a different splice variant is represented by the clone and the Ensembl sequence. By comparison of predicted sequences with experimentally derived, complete sequences we found that in most cases the prediction is correct (data not shown). Predicted insert sequences were generated for 1,133 clones and the corresponding putative sequences of expression products were calculated. For the remaining sequences, the quality of the experimental sequence was not sufficient, or the alignment to the Ensembl transcript suggested that the clone represents a different splice form. The molecular masses derived from the predicted protein sequences were compared to the sizes of proteins expressed in the corresponding clones. Only clones with inserts in the correct reading frame were considered. As shown in Figure [3](#F3){ref-type="fig"}, there is a correlation between the experimental and predicted molecular masses. The correlation is better for clones that express with moderate or high yield (correlation coefficient 0.55, Figure [3a](#F3){ref-type="fig"}) than for clones with weak/doubtful expression (correlation coefficient 0.33, Figure [3b](#F3){ref-type="fig"}). For those clones, where the observed and predicted molecular mass of the expression product match, it can be assumed that the predicted protein sequence is correct to a large extent, and that the clone indeed expresses the expected protein. For clones of interest, this assumption should be verified by sequencing the complete cDNA insert. For other clones, either the sequence was not predicted correctly, because of alternative splicing, for example, or the observed expression product does not correspond the cloned cDNA, because the insert sequence is not expressed completely or because the expression product is degraded within the E. colicells. Public database --------------- The results of our protein expression screening and DNA sequencing of the hEx1 cDNA library are publicly available \[[@B17]\]. The corresponding clones are distributed by the RZPD German Resource Centre \[[@B18]\]. A web interface allows for retrieval of sequence and protein expression data (Figure [4](#F4){ref-type="fig"}). Users can download DNA sequence raw data (chromatograms) and view detailed descriptions of protein expression experiments, including images of SDS-PAGE analyses. Furthermore, users can search for genes and proteins by name, symbol or accession number and display lists of all genes corresponding to clones in the database. These lists can be filtered to display only genes corresponding to full-ORF clones or clones with certain expression properties. Selection of clones and protein preparation for structural analysis ------------------------------------------------------------------- Clones expressing soluble recombinant protein and containing full-ORF inserts were selected for the structural analysis pipeline of the Protein Structure Factory \[[@B2]\]. Clone sequences were matched to the transcript sequences in the Ensembl database. The corresponding Ensembl protein sequences were compared to the protein sequences of the PDB database, using BLASTP \[[@B19],[@B20]\]. Target proteins with known structures were excluded. Specifically, only target sequences were selected with 80% or less sequence identity to PDB entries or with no match to PDB over at least 50 amino acids and at least 10% of the sequence length. One hundred and sixty-three hEx1 clones expressing target proteins with sufficient yield and homogeneity remained after applying these criteria. For preparation of proteins without additional residues such as the His-tag, ORFs were subcloned into the vector pQTEV. This vector allows expression of His-tag fusion proteins and subsequent tag removal by specific protease cleavage using tobacco etch virus (TEV) protease. Of the selected cDNAs, 110 were subcloned into pQTEV, of which 48 were selected for large-scale protein production. A total of 17 of the 48 proteins could be expressed and purified in sufficient yield and quality for protein crystallization. The volume of cultures, grown either in shaker flasks or fermenters, varied between 1 and 5 liters. Protein yields varied from 1.5 to 38 mg/liter of culture volume. Following cell lysis, His-tag fusion proteins were captured by metal chelate affinity chromatography. The His-tag was removed proteolytically and proteins were further purified by ion-exchange and size-exclusion chromatography. The proteins were characterized and prepared for crystallization trials using biophysical methods. A summary of a typical preparation for each clone, and the preparation and characterization data is given in Table [1](#T1){ref-type="table"}. The protein preparations were tested to see whether they were free of aggregates. For 10 of the 17 proteins, this was proven by dynamic light scattering (DLS) analysis. To determine the thermal stabilities, denaturation temperatures (T~m~) were measured by differential scanning calorimetry (DSC). With one exception, all proteins that were free of aggregation showed high T~m~values, of 49-60°C, at pH 7.0 (Table [1](#T1){ref-type="table"}). So far, the structures of gankyrin (PDB 1QYM), aortic preferentially expressed protein 1 and prolidase (unpublished data) have been solved by the Protein Structure Factory as a result of the approach described here. Discussion ========== The expression of soluble recombinant protein is still a bottleneck for functional and structural genomics projects studying human proteins. We demonstrate here a method for generating and characterizing a large set of expression clones for human proteins from a cDNA library, yielding a pre-selection of clones for large-scale expression. By matching clone sequences to the Ensembl database, it was shown that expression clones with soluble products were found for 1,509 human proteins corresponding to 1,105 distinct genes. To cover a larger set of proteins with our approach, additional libraries from different tissues and developmental stages could be used. It was found that 36% of expression clones are full-ORF clones expressing complete human proteins, while the remaining clones express carboxy-terminal fragments. It should be noted that because the Ensembl database is generated automatically and start codon positions are still unknown for many human transcripts, this number is inaccurate and will probably be higher. Future releases of Ensembl will benefit from the ongoing efforts to generate and annotate human full-length cDNA sequences \[[@B21]\], and the information on ORF start positions should improve accordingly. Thre are several reasons for the presence of clones expressing carboxy-terminal fragments. A certain proportion of incomplete inserts is a common feature of cDNA libraries constructed by the cloning technique used here. Furthermore, full-ORF clones containing parts of the 5\'-untranslated region (UTR) are not detected in our expression screen if the UTR contains stop codons. The fact that smaller proteins or fragments are often expressed better than very large proteins in E. colicould be another reason why many clones expressing carboxy-terminal fragments were obtained. Full-ORF clones are generally required for determination of protein structures. However, carboxy-terminal fragments can be interesting for other applications, such as structural analysis of the domain by NMR spectroscopy. As an example of the application of the characterized clone library, we show the selection of clones for structure analysis. The high-throughput screening for expression clones took about a year, while the work on the 163 selected proteins is still in progress and additional proteins are being purified. From the 17 protein preparations, three new protein structures were solved. In conclusion, a systematic screening approach for E. coliexpression clones of human proteins is described here. Using this approach, a public resource of 2,746 clones was created that allows functional genomics projects to select clones and express human proteins of interest. Materials and methods ===================== Sequence analysis and database ------------------------------ cDNA sequences have been submitted to the dbEST database and are available under the accession numbers CD579165-CD580594. Clone DNA sequences were matched to transcript sequences of the Ensembl database, release 20.34c, using the program cross\_match, version 0.990329, of the swat/cross\_match/phrap package \[[@B16]\]. Protein sequences were compared with BlastP \[[@B19]\], version 2.0a19MP-WashU (Warren R. Gish, unpublished work). A database was created to store the results of the protein expression and purification experiments as well as clone sequence data. The Oracle database management system 8.1.6 was used. A web-based front end including search functionality was developed, using the Java programming language. Determination of reading frames ------------------------------- The reading frame of a cDNA insert was determined using the following formula: \\|c~ce,start~ - (c~cv,end~ + l - v~cv.end~) + o - e~ce,end~\\| mod 3, where lis the length of the vector pQE30NST (3,494 bp). In an alignment of a vector and clone sequence, c~cv,end~and ν~cv,end~denote the positions of the end of the matched region on the clone and vector sequence, respectively. Likewise, c~ce,start~and e~ce,start~denote the start positions of the match of clone and Ensembl sequence. ois the start position of the ORF on the Ensembl transcript sequence. For clones that are in-frame to the vector-encoded start codon and His-tag, the formula returns 0. Predicted clone insert sequences were generated from experimental tag sequences and Ensembl transcript sequences by the Perl program seqjoin. seqjoin uses alignments generated by cross\_match to generate combined sequences. It does not generate output for alignments that indicate alternative splicing. The program and documentation are publicly available online \[[@B22]\]. Subcloning of cDNA fragments into pQTEV --------------------------------------- ORFs were PCR amplified from hEx1 cDNA clones using gene-specific primers. Primers were automatically designed using a Perl script that is available on request. Primer length was adjusted to obtain a uniform T~m~of 60-65°C and sense and antisense primers were equipped with BamHI and NotI sites, respectively. For ORFs containing these sites, alternative enzymes producing compatible overhangs were used (BglII, Eco31I or Esp3I). PCR products were cloned into the vector pQTEV (GenBank AY243506). A pipetting robot and microplates were used for PCR setup, restriction digest and DNA purification steps. The resulting plasmid was introduced into E. coliSCS1 cells carrying the pSE111 helper plasmid. pSE111 provides resistance to 15 μg/ml kanamycin and carries the lacIQrepressor and the argUgene for the arginine tRNA that recognizes the rare codons AGG and AGA. The low abundance of this tRNA is especially critical when expressing eukaryotic genes in E. coli\[[@B23]\]. The resulting clones as well as hEx1 library clones are available from the RZPD German Resource Center for Genome Research GmbH (Table [1](#T1){ref-type="table"}). Protein expression in 96-well plates ------------------------------------ Protein expression was performed as described \[[@B14]\]. The hEx1 library is stored frozen at -80°C in 384-well microtiter plates (Genetix, X7001) in several copies. Plates were thawed at room temperature, and 100 μl cultures (2× YT supplemented with 2% glucose, 100 μg/ml ampicillin and 15 μg/ml kanamycin) in 96-well deep-well plates were inoculated with steel replicators and grown over night at 37°C with rigorous shaking (\> 300 rpm). Nine hundred microliters of pre-warmed SB medium supplemented with antibiotics was added, and cultures were grown for 3 h at 37°C, followed by induction of protein expression for 3 h by addition of 1 mM isopropyl-beta-D-thiogalactopyranoside (IPTG) (final concentration). Cells were harvested by centrifugation at 4°C at 2,000gfor 10 min and frozen at -80°C. Protein purification in 96-well format -------------------------------------- Proteins were purified via metal chelate affinity chromatography in a 96-well format. We used an automated procedure on a pipetting robot \[[@B14]\] or a corresponding manual method. According to the manual method, cells were thawed and resuspended in 100 μl lysis buffer (50 mM Tris-HCl pH 8.0, 0.3 M NaCl, 0.1 mM EDTA) by vortexing, followed by addition 2 mg/ml lysozyme and 0.5% Brij 58 in 25 μl lysis buffer. Cells were lysed for 30 min on ice and nucleic acids were degraded by addition of 25 μl of 10 mM MgCl~2~, 0.1 U/μl Benzonase gradeII (Merck) in 50 mM Tris-HCl pH 8.0, brief vortexing and incubation at room temperature for 30 min. An aliquot was collected for SDS-PAGE analysis (whole cellular proteins). Cellular debris was pelleted by centrifugation of the plates at 6,200 rpm for 30 min. Aliquots of the supernatants were collected (soluble cellular protein). Supernatants were transferred to a filter plate (Millipore Multiscreen MADVN6550) and were filtered on a vacuum manifold. Filtrates were collected in a second filter plate. Imidazole was added to 10 mM, and 25 μl of 20% (v/v) Ni-NTA agarose (Qiagen) equilibrated in 50 mM Tris-HCl pH 8.0. Plates were shaken at room temperature for 30 min, followed by removal of cell lysates on the vacuum manifold. The agarose beads were washed three times by shaking in 200 μl wash buffer (50 mM Tris-HCl pH 8.0, 0.3 M NaCl, 20 mM imidazole). Upon complete removal of liquid from the plate, proteins were eluted by addition of 25 μl wash buffer containing 250 mM imidazole. Eluates were collected in a 96-well plate by brief centrifugation. Seven microliters of the eluates and 3.5 μl of the whole and soluble cellular extracts were analyzed by SDS-PAGE (15% polyacrylamide) and Coomassie staining. Large-scale protein production and biophysical characterization --------------------------------------------------------------- Proteins were expressed, purified, concentrated and analyzed as described \[[@B24]\]. Cells were grown in SB media (see above) containing 50 mg/ml ampicillin and 10 mg/ml kanamycin in 5 l baffle shaker flasks in 2 l volumes or in a 5 l fermenter to a cell density of A~600~of 1.5 and protein expression was induced by addition of 1 mM IPTG for 4 h. The optimal expression temperature was determined in small-scale experiments beforehand (28-37°C). Cells were pelleted by centrifugation and resuspended in a threefold volume of 20 mM Tris-HCl pH 7.4, 300 mM NaCl, 10 mM imidazole, 5 mM 2-mercaptoethanol, 1 mM PMSF, a protease inhibitor cocktail tablet (EDTA-free, Roche) and 500 units Benzonase (Merck). Cells were lysed by treatment with lysozyme and sonification, followed by centrifugation (23,000g, 45 min) and filtration through a 0.22-μm syringe filter. Proteins were applied to a metal chelate chromatography using a Ni-POROS20-column (Applied Biosystems) or a TALON column (Clontech). After washing with 20 mM Tris pH 7.4, 150 mM NaCl, 10 mM imidazole, the protein was eluted with 250 mM imidazole in the same buffer and eluates were supplemented with 2 mM dithiothreitol and 1 mM EDTA. The His-tag was removed by incubation with TEV protease (molar ratio 1:40 protease:substrate) at 4°C overnight. Proteins were diluted fivefold and depending on the theoretical pI of the protein, anion or cation exchange chromatography was performed. Proteins were further purified by gel filtration on a Superose 12 16/50 column (Amersham Biosciences). Protein concentrations were determined from the absorbance at 280 nm using the extinction coefficient calculated from the amino acid sequence \[[@B25]\]. Absorbance was corrected for stray light according to the light scattering theory (Tyndall effect, I~(s)~\~ λ^-4^) with the assumption that no absorption due to protein chromophores occurs above 320 nm \[[@B26]\]. Purified protein concentrations were in the range of 0.2-1 mg/ml. DLS measurements were carried out at room temperature, using the Spectroscatter 201 (660 nm laserdiode, 30 mW, scattering angle 90°, PMT detector, 400 nsec to 30 sec correlator, quasi-logarithmic arranged channels, RiNA, Berlin, Germany). The samples were centrifuged (20,800g, 3 min, 4°C) and measured in a 1.5 × 1.5 mm cuvette (Hellma, Müllheim, Germany) for 20 sec. The instrument software allows us to judge the autocorrelation function and deduce the dispersity, that is, the distribution N(Rh), of particles according to their hydrodynamic radius. Protein samples were judged \'free of aggregation\' when a single peak indicated a monomodal distribution. DSC measurements were performed at a rate of 1 K/min using an automated capDSC calorimeter (MicroCal, LLC, Northampton, MA). Proteins were diluted at least 20-fold in a buffer of temperature-independent pH (20 mM Na/K phosphate pH 7.0, 150 mM NaCl). The resulting scans were baseline-corrected and T~m~values were calculated using the instrument software (MicroCal Origin, vers. 7.0). cDNA sequencing --------------- cDNA inserts were PCR-amplified using primers pQE65 (TGAGCGGATA ACAATTTCAC ACAG) and pQE276 (GGCAACCGAG CGTTCTGAAC), annealing temperature 65°C. PCR products were tag-sequenced using primer pQE65. Additional data files ===================== Additional data file [1](#s1){ref-type="supplementary-material"}, available with the online version of this paper, is a tab-delimited text file listing information on hEx1 clones with inserts in the correct reading frame, giving their clone ID, Ensembl transcript ID, experimental and predicted expression product size, expression strength. Supplementary Material ====================== ::: {.caption} ###### Additional data file 1 A a tab-delimited text file listing information on hEx1 clones with inserts in the correct reading frame, giving their clone ID, Ensembl transcript ID, experimental and predicted expression product size, expression strength ::: ::: {.caption} ###### Click here for additional data file ::: Acknowledgements ================ We thank Anja Koch for performing DLS and DSC measurements and Thomas Grund and Dinh-Trung Pham for protein preparation. We are grateful to Martin Strödicke and Erich Wanker for exchanging cDNA clone sequence data. This work was funded by the German Federal Ministry of Education and Research (BMBF) through the Leitprojektverbund Proteinstrukturfabrik and through the grant Development of Platform Technologies for Functional Proteome Analysis - Application to Human Brain (031U102D). Support by the Berlin Senate and the European Fund for Regional Development (EFRE) is also gratefully acknowledged. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flow chart of hEx1 library analysis. ::: ![](gb-2004-5-9-r71-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Results of high-throughput protein purification visualised by SDS-PAGE (15% polyacrylamide) and Coomassie staining. Expression products of six hEx1 cDNA clones (1-6) are shown. W, whole cellular protein extracts; S, soluble protein cellular extracts; E, protein purified under non-denaturing conditions. M, calibration with molecular mass standards. The arrow indicates the position of lysozyme, which was added for cell lysis. ::: ![](gb-2004-5-9-r71-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Comparison of experimental and predicted molecular masses of expression products. The correlation of predicted and experimental protein masses is shown. Dots close to the diagonals have expression products of a size corresponding to the predicted size. (a)Clones with moderate or strong protein expression. (b)Clones with weak or doubtful protein expression. For clones with at least moderate expression, a good correlation of predicted and experimental molecular mass of the expression product is visible. r, correlation coefficient. ::: ![](gb-2004-5-9-r71-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Database web interface. The search result for the term \'adenylate\' is shown. ::: ![](gb-2004-5-9-r71-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Protein preparation and characterization ::: hEx1 clone\* pQTEV subclone^†^ GenBank accession number GenBank protein name Protein yield per culture volume^‡^(mg/l) Free of aggregation^§^(DLS) Melting temperature^¶^(°C) -------------- ------------------- -------------------------- -------------------------------------- ------------------------------------------- ----------------------------- ---------------------------- P08596 758H0126 AAH01214 CGI-68 1.5 Yes 60.3 (1) H05552 758H0426 AAF76210 DRG-1 1.5 No 54.8 ± 0.8 (2) E10541 758G1126 AAH07250 Coatomer protein complex, subunit ε 1.7 No 50.8 (1) E12601 758H1026 CAC37950 HCC-1 7 No No peak E09507 758C041 AAB59004 ARA9 4 Yes 54.7 ± 0.4 (2) I07544 250H071 AAH05289 Thioredoxin-like 7 Yes 56.6 ± 2.9 (3) N13557 250B082 AAH06346 Aortic preferentially expressed p. 1 18 Yes 55.5 ± 2.2 (4) J17573 250A012 AAA58682 Transformation-sensitive protein 19 Yes 51.1 (1) G03570 250A062 AAH11960 Gankyrin 3 Yes No data L02591 250B092 AAH12995 Protein for MGC:3363 8 No 41.4 (1) C16543 250B062 AAH04430 Unknown 2 Yes 49.2 (1) H07543 250B112 BAA78534 Adenylate kinase 1 38 Yes 51.6 (1) H14508 250F062 BAA04802 HUM22SM 38 No data 53.4 (1) E07518 250A053 AAF87955 NBP 7 Yes 32.0 (1) H08589 250A073 AAH07873 Unknown 3 No No peak A23602 250C112 CAB53072 Microtubule-associated protein 4 No 58.0 (1) M05547 250H122 AAH15027 Prolidase 10 No 50.8 ± 1.3 (3) \*RZPD clone identifiers without prefix \'MPMGp800\'. ^†^Clone in vector pQTEV used for protein production, RZPD clone identifiers without prefix \'PSFEp\'. ^‡^Protein yield of one typical protein preparation. ^§^As determined from measurements of purified protein preparations of 0.2-1 mg/ml. ^¶^Melting temperature determined by DSC. The number of independent measurements is indicated in brackets. Proteins for which no peak was observed were considered to be unfolded. :::	PubMed Central	2024-06-05T03:55:48.540097	2004-8-17	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522878/", "journal": "Genome Biol. 2004 Aug 17; 5(9):R71", "authors": [ { "first": "Konrad", "last": "Büssow" }, { "first": "Claudia", "last": "Quedenau" }, { "first": "Volker", "last": "Sievert" }, { "first": "Janett", "last": "Tischer" }, { "first": "Christoph", "last": "Scheich" }, { "first": "Harald", "last": "Seitz" }, { "first": "Brigitte", "last": "Hieke" }, { "first": "Frank H", "last": "Niesen" }, { "first": "Frank", "last": "Götz" }, { "first": "Ulrich", "last": "Harttig" }, { "first": "Hans", "last": "Lehrach" } ] }
PMC522879	Background ========== The complete genome sequences of various eukaryotic model organisms such as Saccharomyces cerevisiae, Caenorhabditis elegans, Drosophila melanogaster, Arabidopsis thalianaand Schizosaccharomyces pombe, have revealed a large number of novel genes of unknown functions. In S. cerevisiae, for example, around 1,800 genes (of the total of around 5,800) encode proteins that so far remain functionally uncharacterized (compilation from SaccharomycesGenome Database (SGD) \[[@B1]\] April 2004). Since the completion of its DNA sequence \[[@B2]\], the genome of S. cerevisiaehas been extensively studied, serving as a test case for novel and important developments in functional genomics. Such developments include transposon-mediated gene inactivation and tagging \[[@B3]\], the analysis of gene-expression networks through partial or complete transcriptome studies \[[@B4]-[@B6]\], two-hybrid screening \[[@B7]-[@B9]\], protein-complex purification \[[@B10],[@B11]\], two-dimensional gel protein identification \[[@B12]\], proteome qualitative analysis by protein microarrays (see review in \[[@B13]\]) and protein abundance measurements after in situgene tagging \[[@B14]\]. Even intergenic regions have been studied using microarray technology to characterize transcription-factor-binding sites and to map replication origins or recombination hotspots \[[@B15],[@B16]\] (see also \[[@B17]\] for a review). Following a large cooperative effort between European and American labs, a nearly complete collection of deletion mutants of all yeast protein-coding genes is now available \[[@B18]-[@B20]\], which offers the possibility of systematically screening numerous phenotypes, including synthetic lethals \[[@B21]-[@B23]\], in search of novel gene functions. As a complement to gene inactivation, phenotypic changes resulting from gene overexpression may also be informative of gene functions. Indeed, in a number of cases, such as genes encoding cytoskeletal proteins or protein kinases and phosphatases, overexpression may lead to a lethal phenotype (see \[[@B24]\] for a review). The overexpression approach is complementary to the loss-of-function approach, as it leads to dominant phenotypes even in the presence of the wild-type gene, thus allowing the study of genes for which no loss-of-function mutants can be obtained. Overexpression of gene fragments can be equivalent to \'dominant negative mutation\' in which the fragment disrupts the activity of the wild-type gene \[[@B25]\]. Overexpression can also activate specific pathways, leading to deleterious phenotypes: examples include genes involved in the yeast pheromone response pathway, such as STE4, STE11and STE12(see \[[@B24],[@B26]\] and references therein). In other cases, specific effects are not known, but the region responsible for toxicity has been identified. For example, lethality upon overexpression of Rap1p depends on the presence of the DNA-binding domain and an adjacent region \[[@B27]\]. In general, however, unless the domain structure of the protein is well understood, one cannot predict which segment(s) of it would act as a dominant mutant when overexpressed. Several yeast cDNA libraries have been screened for lethal or impaired growth phenotypes upon overexpression under the control of the GAL1or GAL10promoters on centromeric or multicopy plasmids \[[@B28]-[@B30]\]. Other libraries of random genomic DNA have also been screened for toxicity upon overexpression from the same promoters \[[@B24],[@B26]\]. Whereas the four earlier studies each identified only a few genes (from 1 to 24 each, making a grand total of 43), Stevenson et al.\[[@B30]\] identified 185 genes (20 of which were shared with earlier work) that cause impaired growth when overexpressed. In the work reported here, we have screened the yeast genome with the aim of characterizing a list of fragments whose overexpression confers growth impairment. To do this, we constructed a yeast genomic library in a multicopy plasmid vector in which transcription is driven by a chimeric tetO-CYC1promoter \[[@B31]\]. Random genomic inserts of a mean size of 700 base-pairs (bp) were overexpressed in yeast as translational fusions using the plasmid-borne initiation codon. Out of around 84,000 clones tested, we have identified the largest collection yet of toxic overexpressed fragments in yeast: 714 showed overexpression-dependent lethality or various degrees of growth impairments, identifying 454 protein-coding genes (91 of which are of unknown functions), and a variety of intergenic or other regions. Results ======= Screening the library of yeast random genomic fragments for toxic phenotypes ---------------------------------------------------------------------------- We have analyzed a total of 84,086 independent yeast transformants, each of which contains a random fragment of the yeast genome placed under the control of a doxycyclin-repressible promoter (Figure [1a,1b](#F1){ref-type="fig"}). Effects on growth or survival were monitored by spotting serial dilutions of the transformants in the presence and absence of doxycyclin (uninduced and overexpression conditions respectively, Figure [1c](#F1){ref-type="fig"}). Phenotypes were recorded using numerical values from 0 to 3 (Figure [2](#F2){ref-type="fig"}): value 3 was assigned to normal growth (similar to non-toxic control), 2 and 1 were assigned to intermediate growth levels (less abundant and/or smaller-sized colonies), and 0 was assigned to complete or almost complete absence of colonies (comparable to the toxic control on the same plate). We have retained 714 clones (0.85% of total) that show impaired growth in overexpression conditions (Table [1](#T1){ref-type="table"}). Among these, 112 also show a slight or severe growth reduction (level 2 for 77 cases, or level 1 for 35 cases, respectively) in unexpressed conditions. Proof that the observed growth defects were caused by the presence of the plasmid rather than an accidental mutation in the clone was directly demonstrated by the recovery of the wild-type phenotype after plasmid loss using selection for resistance to 5-fluoroorotic acid (5-FOA) (Figure [2](#F2){ref-type="fig"}). Identification of the genomic inserts conferring toxic phenotypes ----------------------------------------------------------------- Inserts of the selected clones were identified by DNA sequencing (Materials and methods). The complete list of inserts is described in Additional file 1 and 2, and results are summarized in Table [1](#T1){ref-type="table"}. A majority of inserts (493, or 69% of total) carry in-frame portions of annotated open reading frames (ORFs), excluding Ty and Y\' ORFs. In addition, a significant number of inserts (162 (23%)) correspond to fragments of ORFs cloned either in antiparallel orientation or out-of-frame with respect to the initiator ATG codon or to intergenic regions. The 59 remaining cases (8% of total) correspond to fragments of transposable elements (17 clones) and subtelomeric Y\' elements (9 clones), to RNA-coding genes (4 clones), and to non-chromosomal replicons such as the 2 mm plasmid and mitochondrial DNA (mtDNA) (29 clones). If any random fragment of the yeast genome were capable of generating a toxic phenotype, in-frame ORF fusions would represent only around 10-12% of the selected inserts (around 70% of the genome correspond to coding regions, and only one frame out of six corresponds to the natural frame). The fact that the toxic inserts correspond principally to in-frame portions of natural ORFs suggests that the coding part of the genome is the most prone to confer toxicity when overexpressed. Analysis of domains within in-frame ORF fragments ------------------------------------------------- The 493 inserts corresponding to in-frame ORF fragments represent 454 distinct annotated ORFs (see Materials and methods), which are randomly distributed throughout the 16 chromosomes of S. cerevisiae(see Additional file 1). In our screening, 32 ORFs were found twice, two ORFs were found three times and one ORF (YHR056c in the CUP1region) was found four times, the cloned fragments being either overlapping (22 ORFs) or non-overlapping (13 ORFs). Mean size of the coding region of inserts is 659 bp. The chosen cloning strategy favors recovery of central-or carboxy-terminal coding parts of the natural yeast genes, whereas the amino-terminal coding regions are rare \[[@B7]\]. In our work, the cloned insert encompasses the entire gene in only six cases (additional file 3, column 20 to 23). In 154 additional cases, the insert corresponds to the carboxy-terminal portion of the natural protein (the stop codon is present). In 10 cases, the inserts start upstream of the natural ATG initiator codons, lengthening the natural peptides by reading in-frame through the untranslated region. Other cases correspond to the central coding region of natural genes. To find possible common characteristics, we have compared between themselves all the peptides encoded by in-frame ORF fragments. BLASTP analysis was combined with detection of characterized conserved domains, of COG patterns (clusters of predicted orthologous groups of proteins \[[@B32]\]), and of transmembrane spans (TMS) to identify toxic inserts similar to each other (see Materials and methods). Out of the 493 in-frame ORF fragments, a total of 170 were divided up into 57 distinct groups of similarity, containing from two to 12 inserts, including overlapping fragments of the same ORF (see Additional file 4). It is expected that several ORFs from a same paralogous gene family are found in a same group. Note that in 16 out of 57 groups, the inserts contain transport-specific domains and/or transmembrane spans. As well as comparing inserts to each other, we also analyzed the totality of the conserved domains present in all peptides encoded by the 493 toxic inserts (see Materials and methods). Characterized domains are found, at least partially, in a total of 281 inserts (see additional file 1 and 3). Of a total of 183 distinct domains, 46 are represented more than once. We have compared the frequency of these 46 domains among the toxic inserts versus their frequency among the 5,803 ORF-encoded proteins of the entire genome (Table [2](#T2){ref-type="table"}). We find that 37 domains are significantly over-represented compared to a random expectation, suggesting that we have screened specific domains. These 37 domains correspond predominantly to various transporter domains (11 cases), such as amino-acid permeases and mitochondrial carrier protein domains. The toxicity of these domains is probably due to the presence of transmembrane spans. Indeed, 132 out of the 493 toxic peptides contain at least two transmembrane spans, including cases where one span is putative (see Materials and methods). Among these, 63 contain three or more predicted spans and 26 have five spans or more. Putative spans were also recognized in 84 other ORF fragments (seven with at least three spans, 15 with two spans, and 62 with one span) (see Additional file 1 and 3). RNA-and DNA-binding domains (nine cases) involved in replication, transcription or translation functions, such as PUF, KH and rrm, are also much more represented than expected (Table [2](#T2){ref-type="table"}). The PUF domain is also involved in recruitment of proteins into a complex that controls mRNA translation (see \[[@B33]\] for review). Other important domains for interactions with polypeptides, phospholipids or small molecules (nine cases) are also over-represented. The WD40 motif, a propeller-like platform for stable or reversible binding of proteins in eukaryotes, has been found in inserts of 12 distinct ORFs (see additional data file 1). The 12 ORFs code for proteins having interactions with other proteins in complexes related to RNA processing or transcription \[[@B10]\], and nine have at least one partner also selected during our screening (see Discussion). Other interacting domains were found, such as dynamin, MRS6, and adaptin\_N domains, which have roles in the dynamics of proteins, membranes and cytoskeleton, and PBD, a small domain which binds small GTPases and inhibits transcription activation. The PH domain, which binds phosphoinositides or other ligands and is involved in signal transduction, was found in inserts of three distinct ORFs involved in different functions: metabolism, cell fate, transcription (see Additional data file 3). Finally, other over-represented domains are related to metabolism and other functions (eight cases), of which several may be involved in interactions with other domains. The serine/threonine protein kinase domain (S\_TKc) is significantly under-represented in our screen. Among the 10 toxic inserts whose cognate genes code for protein kinases (PK), only four contain this domain (Additional data file 3). In these four cases, the S\_TKc domain is either truncated (Additional data file 4), or flanked by a coiled-coil region and/or a low-complexity segment. Two other inserts contain the PBD (and PH) domains, and the four remaining inserts contain no characterized domain to date. As it is known that overexpression of some protein kinases is deleterious for cells (see \[[@B24]\] and references therein), our results suggest that a domain different from the catalytic domain is responsible for the toxicity of these proteins, and that the fragments selected in our screen have a role in binding ligands such as substrates or regulators of protein kinase activity, or of proteins involved in the signaling cascades. Three other genes coding for protein kinases of the phosphatidylinositol 3-kinase (PI kinase) family are also represented in our screen by four toxic inserts, none of which contained the kinase domain (see Discussion). The remaining 137 domains (out of 183) were found only once each. Many correspond to functional categories described above, such as transport, metabolism, and interactions with nucleotides, other proteins or other ligands. Seven domains associated with ubiquitination functions were also found (see Additional data file 3 and 5). Several of the domains encountered have also been isolated as mammalian genetic suppressor elements (GSEs), which are cDNA fragments that inhibit cell growth (see \[[@B34]\] and references therein). In addition to the domains described above, we found toxic inserts coding for natural peptides without recognizable domains but containing regions of low complexity (56 cases). A number of these peptides are highly charged, either negatively or positively (see Additional data file 3). Such charged peptides might interact in an artifactual way with other charged domains of proteins or nucleic acids or with small molecules. Interestingly, the prion-like (Q+N)-rich domain was found in eight of the natural peptides having low-complexity regions. Nature of the selected genes ---------------------------- We have seen above that 493/714 toxic inserts are in-frame fragments of protein-coding genes. The complete list of the 454 genes corresponding to these toxic inserts is given in Additional dat file 1 and 2. Their sizes range between 282 bp and 14,733 bp. The mean size of this distribution is 2,401 bp (standard deviation (SD) 1,671 bp), to be compared with a mean size of 1,444 bp (SD 1,094 bp) for the entire set of 5,803 ORFs of the yeast genome. The bias towards longer ORFs is expected from our cloning strategy (see above). Note that the 35 ORFs that we found more than once are nearly randomly distributed in various size classes. We examined the distribution of these genes according to different criteria, such as function, subcellular localization, viability and phylogeny (Table [3](#T3){ref-type="table"}) and compared it to the distribution of the genes of S. cerevisiae. Among the 454 ORFs identified, 91 are unclassified, and function is not yet clear for six others (see Additional data file 3). The remaining ORFs represent a variety of functional classes (Table [3](#T3){ref-type="table"}). Distribution of the 454 ORFs shows statistically significant deviations for eight out of the 15 functional classes, taking into account biases due to mean size of genes in each class. Globally, there is a deficit of genes involved in protein synthesis and of unclassified genes, and an excess of genes involved in transport facilitation and cellular transport (echoing the fact that we found many inserts containing transporter domains and transmembrane spans), in cell fate, in transcription and, to a lesser extent, in cell cycle/DNA processing and in homeostasis (regulation of/interaction with the environment). As seen above, many toxic inserts contain multiple predicted TMS. Such inserts correspond most often to genes coding for transporters or for non-transporter membrane proteins \[[@B35]\]. We have selected a total of 96 transporters (see Additional data file 3) of which 18 belong to the class of putative uncharacterized transporters, whose toxic inserts contain several TMS. Fourteen others belong to the class of transporters of unknown classification, including 13 genes of the nuclear-pore complex family, whereas there is a total of 58 genes in this family in the whole genome. On the other hand, 24 genes coding for non-transporter membrane proteins were also selected. Taken together, 120 transporters and non-transporter membrane proteins are represented in our screen, twice as many as expected (61 expected), as 782/5,803 ORFs are known or predicted as coding for such proteins \[[@B35]\]. The distribution of the proteins encoded by these genes in the cell is strongly biased in favour of the plasma membrane and against the cytoplasm, and, to a lesser extent, in favour of nucleus and cytoskeleton (Table [3](#T3){ref-type="table"}). Although the majority of inserts originate from non-essential genes, we have found 96 essential genes (21%) among the selected ORFs. This is a significantly higher percentage than in the whole genome, where 939/5,803 genes (16.2%) are essential (Table [3](#T3){ref-type="table"}). Using the classification from Malpertuy et al.\[[@B36]\] and additional updating (Génolevures \[[@B37]\]), we find that the majority of genes yielding toxic fragments in this work are conserved (336/454 (74%)) between S. cerevisiaeand other sequenced organisms, whereas 106 (23%) are ascomycete-specific and 10 (2.2%) are orphan genes. This distribution is significantly different from the distribution among the 5,803 genes of S. cerevisiae, where 64% of protein-coding genes are conserved (see Table [3](#T3){ref-type="table"}). The under-representation of orphan genes in our screen is already apparent in the under-representation of functionally unclassified genes, as a high rate of orphans of the whole genome (79%) are also unclassified (data from Génolevures \[[@B37]\] and Munich Information Center for Protein Sequences (MIPS) \[[@B38]\]). Toxicity of entire genes versus ORF fragments --------------------------------------------- To compare the phenotypes conferred by overexpression of the entire gene and of the gene fragment, we have cloned the cognate entire genes of 13 in-frame toxic inserts into the vector pCMha191 (see Materials and methods). One criterion for the choice of the genes was the absence of a mutant phenotype of the corresponding gene disruption at the time this work was started, except for the NOP4gene whose disruption is lethal. Six of these genes are singletons; three others have a paralog already known as toxic upon overexpression. Six out of the 13 still have no known function to date (Table [4](#T4){ref-type="table"}). Expression at the protein level of both entire gene and gene fragment was verified by western-blot analysis, using an anti-hemagglutinin (HA) antibody (data not shown). As seen in Table [4](#T4){ref-type="table"} and Figure [3](#F3){ref-type="fig"}, we found that overexpression of 10 genes was as toxic or more toxic than overexpression of the gene fragments. One gene, YGR149w, was less toxic in its entire version than in the truncated form, which was weakly toxic. Finally, we found that two genes, YML128c/MSC1and YDL112w/TRM3, showed no toxicity when overexpressed, whereas the cloned inserts were strongly toxic. In these two cases, the immunolocalization of overexpressed products was examined, and the cytoplasmic localization of the fragment agreed with the location of the natural gene product (data not shown), indicating that the toxic effect is not the result of mislocalization of the overexpressed fragment. The gene MSC1had already been screened \[[@B24]\] as a toxic fragment in overexpression conditions, the region concerned being the same as in our screening. This gene has low similarity to a stress protein of Schizosaccharomyces pombeand has a role in meiotic recombination. The TRM3gene contains a carboxy-terminal domain responsible for tRNA methyltransferase activity \[[@B39]\], which is absent from our insert. The protein is a member of a complex probably involved in signaling \[[@B10]\]. Analysis of other fragments --------------------------- Additional data file 2 analyzes the 221 other toxic inserts which do not correspond to in-frame fragments of annotated ORFs. Sixty-eight inserts correspond to natural ORF fragments cloned in an antiparallel orientation, most of them being entirely included within the ORF sequence (47 cases), the others overlapping the intergenic upstream region of the natural ORF (17 cases) and sometimes the next gene as well (four cases). Their toxicity can result either from the overexpression of an antisense RNA or from the overexpression of a toxic artificial peptide encoded by a fortuitous ORF. Several arguments favor the second hypothesis. First, short ORFs longer than 24 codons (maximum observed 250 codons), and in-frame with the start codon of the cloning vector, are observed in 53 cases (78% of the total). A number of those artificial ORFs are due to the \'mirror\' effect produced by codon-biased natural ORFs \[[@B40],[@B41]\]. But the fact that they are observed more than one-third of the time suggests a positive selection for toxic artificial peptides. Second, antiparallel ORF fragments do not correspond to a majority of essential genes, as might be expected from antisense RNA inhibition. Third, we have directly verified, for two inserts recloned in the same vector, that addition of a stop codon that blocks translation of the artificial ORF also suppresses toxicity (see Additional data file 9). Even if this concerns only two cases, we have no direct results indicating the existence of antisense RNA molecules that could block expression of essential genes. Fifty-three additional inserts correspond to natural ORF fragments cloned out-of-frame with respect to the plasmid-borne ATG codon, of which only 12 code for artificial ORFs longer than 24 codons (see Table [1](#T1){ref-type="table"} and Additional data file 2). Intergenic regions are represented by 41 inserts, of which 27 (65% of total) code for short artificial ORFs. In total, short artificial peptides may be encoded by 92 out of the 162 inserts described above. Comparison of the 92 peptides between themselves reveals several low-complexity sequences (see Additional data file 2), mostly encoded by antiparallel ORF fragments whose direct amino-acid sequence is itself of low complexity. Comparison with the proteins of S. cerevisiaeand of all available sequenced organisms compiled in our internal database (GPROTEOME3, see Materials and methods) reveals no significant similarity. None of these artificial ORFs corresponds to the 137 new annotated yeast genes of Kumar et al.\[[@B42]\], to the 62 new genes of Oshiro et al.\[[@B43]\] or to the 84 genes of Kessler et al.\[[@B44]\]. Even though we have no evidence for antisense RNA activity, we cannot exclude a toxic effect due to the overexpressed transcript itself. Among the 59 remaining inserts, 17 belong to Ty elements, 10 of which are in-frame ORF fragments corresponding to TyB only (two of them containing the carboxy-terminal part of the rve domain (integrase core)), whereas all antisense fragments (three inserts) correspond to TyA. Y\' elements, which are present in 20 copies in the genome, are represented by nine inserts, all coding for highly basic or acidic peptides (of which three are in-frame fragments of natural ORFs) which contain repeats of amino acids or motifs, and confer a strongly toxic effect (see Additional data file 2). Considering that these inserts are toxic, their observed number is not different from that expected from the size and number of Y\' in the genome. Four inserts from yeast chromosome XII are fragments of genes coding for 18S or 25S RNA, two inserts being cloned in the sense orientation. The 2 mm plasmid is represented by 17 fragments, 10 of which are in-frame fragments of ORFs coding for REP1, REP2and FLP1. The seven other inserts are out-of-frame or antisense fragments of FLP1, or fragments of intergenic regions, all (except two) coding for artificial ORFs. Finally, mtDNA is represented by 12 fragments, mostly corresponding to intergenic regions on the minus strand of the chromosome. Artificial peptides highly enriched in the amino acids tyrosine (Y), isoleucine (I), and lysine (K) are encoded by 10 out of the 12 mitochondrial inserts. Discussion ========== The general fitness of living organisms largely depends on a harmonious equilibrium between the various cellular components and on their capacity to maintain homeostasis. The intricate circuitries that regulate gene expression form the basis of these properties, and massive deregulation of single components may result in flagrant phenotypic defects leading to serious growth impairment or even cell death. Our large-scale screening of the yeast genome using random genomic fragments resulted in a collection of several hundreds of inserts showing toxic effects on cell survival or growth when overexpressed. These toxic effects are expected to result from several distinct molecular situations that have been encountered at various frequencies in our experiments. Of the total of 714 toxic inserts studied, a majority (69%) correspond to the overexpression of fragments originating from natural protein-coding genes (454 genes were identified in total). But, interestingly, a large minority (23%) correspond to noncoding DNA fragments. The remaining cases (less than 10% of the total) correspond to fragments of Ty or Y\' elements, of the 2 μm plasmid or of mtDNA which, after analysis, can be attributed to one of the two previous categories. Toxic fragments of natural gene products are interesting to consider with respect to the functions of the corresponding genes. But the second category may be even more promising in that it offers us a description of DNA sequences that cannot be overexpressed in a cell without a deleterious effect. The toxicity of coding fragments may result from the imbalance between products of tightly controlled genes, or from the titration of active complexes by the presence of truncated proteins and/or isolated domains. In addition, nonspecific effects might also exist, for example, as a result of an abnormal intracellular localization of an artificially overabundant peptide or protein. We did not attempt to distinguish experimentally between these possibilities for all the coding inserts isolated in this work. Taking into account only specific effects, in the limited number of cases in which the entire gene corresponding to a toxic insert was cloned in the overexpression vector (see Results), we verified that toxicity was due, in most cases, to the disruption of the precise dosage of an essential cellular component (the entire protein is also toxic when overexpressed) and, in some cases, to the titration effect exerted by the incomplete fragment of the natural protein (the entire protein is not toxic when overexpressed). A few examples where the domain responsible for toxicity upon overexpression is known can be found in the literature. In the case of TOR1and TOR2genes, toxicity is specific to a central domain of the proteins distinct from their carboxy-terminal protein kinase domain; overexpression of the entire gene has no effect, and can even cure the negative effect of the overexpressed domain \[[@B45]\]. Alarcon et al.\[[@B45]\] have proposed that Tor proteins could serve as a scaffold on which to assemble other proteins for appropriate interaction with the kinase domain. Our results agree with this hypothesis, as four out of the five yeast genes belonging to the conserved family of PI kinase-related protein kinases - TOR1, TOR2, TEL1and TRA1- were selected in this work, all represented by inserts of the central region of these proteins (Figure [4](#F4){ref-type="fig"} and Additional data file 3). In mammalian cells, overexpression of such fragments of ATM, a homolog of TEL1, also has a negative effect \[[@B46]\]. In other cases in which overexpression of the entire gene is toxic, certain domains responsible for the toxicity have been mapped, for example the Myb DNA-binding domain of RAP1(see Background), the ZnF C3H1 domain of CTH1\[[@B47]\] and the bZIP domain of GCN4\[[@B48]\]. All these DNA-binding domains were significantly over-represented in our screen (Table [2](#T2){ref-type="table"}). Even in the absence of precise mapping of the toxic domain present in our clones, we were able to explore the nature of the domains found in each insert. Our experiment has shown a bias towards domains corresponding to transport functions and to various interactions (Table [2](#T2){ref-type="table"}). As mentioned in Results, the toxic effect of transport-specific domains may be due to the presence of corresponding TMS. As our results also showed a bias towards a number of interaction domains, we have examined the known interactions of the proteins encoded by the 454 genes found in this screen (see Materials and methods). Genetic interactions were also considered, excluding the coexpression results obtained in microarray experiments. It appears that 88.3% of our genes (401/454, of which 70 are of unknown function) code for proteins which have known genetic or physical interactions, or are members of complexes (see Additional data file 3). Moreover, for 60% of these (242/401), at least one of their known partners is also found in our screen (see Additional data file 6 and 7). Among the 53 genes having no known interactions, 24 correspond to transporter or membrane proteins (see Additional data file 3). The biases we have observed show little overlap with previous screenings of S. cerevisiae, which had previously identified a total 231 genes or gene fragments that were toxic when overexpressed \[[@B24],[@B26],[@B28]-[@B30],[@B49]\]. Among the 185 genes of Stevenson et al.\[[@B30]\], those involved in protein synthesis are represented twice as frequently as in the whole genome, whereas they are twice less frequent in our own experiment. In contrast, genes involved in transport facilitation and interactions with the environment were not over-represented in the Stevenson et al.experiment. Common biases are, however, observed in favor of transcription, cell-cycle and cellular transport genes. Overall, only 33 of our 454 ORFs were previously identified by the previous authors (the total rises to 78 if one considers individual gene studies). Twenty-five other genes from the previous screenings not found here are members of paralogous gene families represented in our work (see Additional data file 3). The limited overlap may result from partial genome coverage. However, by screening 84,086 clones (a coverage of around 4.5 genome equivalents), we must have encountered a total of 4,677 ORFs, each being represented 1.6 times as an ORF fragment (see Materials and methods). We have thus screened for toxicity around 80% of the natural yeast ORFs. But the limited overlap of results may also be explained by the experimental bias introduced by each technique. The previous experiments were mostly based on cDNA cloning, which favors short and highly expressed genes, whereas our genomic library favors large ORFs (mean size 800 ± 557 codons per ORF) and has no expression bias. In addition, the largest previous experiment \[[@B30]\] was done using centromeric plasmids and a galactose promoter as opposed to our multicopy vectors. Furthermore, our serial dilution drop assay is probably more sensitive to growth alteration than the replica techniques previously used. Finally, previous overexpression experiments relied on changing the nutrient composition of the growth medium (galactose vs glucose) whereas our experimental set-up relied on the presence/absence of a drug in a medium of the same nutritional composition. The finding of a large minority of toxic inserts corresponding to noncoding DNA is puzzling. Indeed, some of the toxic inserts originate from annotated but questionable ORFs, and some originate from antisense or intergenic fragments which can artificially be translated into small ORFs. None of these peptides has recognizable characterized domains, but many of them are charged, mostly positively (see Additional data file 2) and some have amino-acid sequences of low complexity. It could be proposed that all these small ORFs represent a reservoir of potentially new gene sequences in the genome. In addition, 100 of the in-frame toxic inserts had no characterized domains and sometimes no predicted secondary structure. These inserts do not contain conserved domains, COGs or TMS, and are not biased in amino-acid composition (see Additional data file 3). They may correspond to domains that have not yet been described, or to domains whose structure has diverged, but another possibility would be that some protein domains are perhaps not structured in a permanent way before evolving towards a structurally functional domain. Interestingly, a significant proportion of the expressed peptides we selected are specific to ascomycetes, or are even true orphan genes that have no known homolog in any other species than S. cerevisiae. A collection of toxic polypeptides, acting as genetic suppressor elements and interfering with major cellular functions, is of interest not only in antifungal research but also as a means of identifying new domains with major physiological roles. Finally, given the large number of inserts encoding very short ORFs (around 70 amino acids in the groups of antiparallel, intergene and out-of-frame fragments, and 80 in total, see Additional data file 2), we cannot exclude the possibility that some transcripts are toxic through hypothetical mechanisms that may include, for example, nonspecific interactions with other cellular or nuclear complexes or through overloading of some component(s) involved in RNA metabolism. Conclusions =========== In a large-scale phenotypic screening of overexpressed random DNA fragments, we selected around 470 genes (including Ty, Y\' and the 2 μm plasmid) whose domains inhibit or impair growth when overexpressed. Many functional categories are represented, transporter proteins being especially over-represented, and genes of unknown function represent one-fifth of our selection. Our approach gave access to genes controlling intracellular and membrane structures, as well as to genes whose deficiency is compensated for by genetic redundancy. Comparable approaches, using efficient phenotyping technology \[[@B50]\] and appropriate screening procedures, could be used for identification of genes involved in specific functions, such as homeostasis and response to stress. We have carried out an analysis of toxic protein domains, pointing out the importance of binding domains and of protein-protein interactions correlated to regulation of cell growth and cell division. This provides a large body of data for targeting more specific studies on the modular construction of proteins and the role of interaction domains in multicomponent assembly of physiological complexes. Finally, in some cases, the deleterious effects in our system of inserts that encode very short ORFs may suggest that overexpression of some transcripts is also toxic for cell growth. Materials and methods ===================== Strains and media ----------------- Total yeast DNA from strain FY1679 (Mata/α, ura3-52/ura3-52, trp1-Δ63/+, leu2-Δ1/+, his3-Δ200/+) \[[@B51]\] was a generous gift of A. Harington. Strains FYBL2-5D (Matα, ura3-Δ851, trp1-Δ63, leu2-Δ1) \[[@B52]\] and FYAT-01 (Matα, ura3-Δ851, trp1-Δ63, leu2-Δ1, his3-Δ200, ade2-661) (A. Thierry, unpublished work) were used for transformations and growth defect screening. All strains are isogenic derivatives of S288C. The yeast genomic library was constructed using Escherichia coliDH10B cells (Electromax DH10B, Gibco-BRL). Yeast cells transformed by pCMha190 recombinants were grown at 30°C on glucose synthetic complete medium lacking uracil (SC - URA) always supplemented with 10 μg/ml doxycycline (Sigma) (uninduced conditions). Phenotypic tests were done on solid medium (12 cm × 12 cm plates) containing 70 ml of SC - URA + 10 μg/ml doxycycline (uninduced conditions) or SC - URA without doxycycline (overexpression conditions). Yeast cells transformed by pCMha191 recombinants were grown at 30°C on SC - tryptophan medium, with or without addition of doxycycline. Plasmid loss was carried out on SC plates containing uracil (50 mg/l) and 0.1% of 5-fluoroorotic acid (5-FOA). Vector construction and cloning ------------------------------- Plasmids pCMha184, pCMha189, and pCMha190 were derived from the centromeric (pCM184, pCM189) or episomal (pCM190) overexpression vectors, containing a tetracycline-regulatable promoter system and URA3(pCM189, pCM190) or TRP1(pCM184) as selection markers \[[@B31]\]. In the original vectors, a 33 bp BamHI-NotI fragment was replaced by a synthetic linker with ends compatible with these sites and introducing an ATG codon followed by an in-frame HA-tag, a BamHI cloning site, and stop codons in the three frames (Figure [1a](#F1){ref-type="fig"}). The episomal pCMha191 vector was derived from pCMha184 (TRP1selection marker) by replacement of the centromere and replication origin with a 2 μm plasmid replication origin. This was PCR-amplified from pCMha190 using primers M1 and M2 (see Additional data file 8) using Pfupolymerase (Stratagene), and ligated to the 5,953 bp EcoRI-BglII fragment of pCMha184. The overexpression system was checked by cloning two short genes, MCM1and AUAI(861 and 285 bp respectively), which are toxic when overexpressed under the control of a GAL1promoter \[[@B29]\]. Both genes were PCR-amplified from yeast genomic DNA (see primers in Additional data file 8), cloned into vectors pCMha189 and pCMha190, and transformed into yeast strain FYAT-01. Only gene MCM1, cloned in the high-copy pCMha190 vector, had a clear and constant toxic effect on yeast growth when overexpressed. We thus decided to build the library into pCMha190 and to choose the MCM1gene as a control for toxic phenotype in overexpression conditions. Thirteen complete genes corresponding to 13 selected toxic inserts (see Results) and the MCM1control gene were cloned into the BamHI digested plasmid pCMha191 (TRP1marker).Genes were PCR-amplified from genomic DNA (see primers in Additional data file 8). For each gene, two independent plasmids were transformed into yeast strain FYBL2-5D. In parallel, the same strain was transformed with the plasmids bearing the corresponding toxic inserts. Two toxic inserts, 156C1 and 57B6, which are antiparallel fragments of YGL039wand YAL062w/GDH3ORFs, were modified by PCR synthesis (see Additional data file 9), then recloned in vivointo pCMha190 using homologous recombination \[[@B53]\] in yeast strain FYBL2-5D. Constructions were verified by sequencing. In parallel, original plasmids extracted from transformed strain FYAT-01 were retransformed into strain FYBL2-5D. Phenotypes in uninduced and overexpression conditions were observed in seven independent transformants in each case. Construction of a random yeast genomic library into pCMha190 ------------------------------------------------------------ The adaptor-based strategy \[[@B7],[@B54]\] was used to prevent self-ligation of the vector and ligation of multiple inserts. Sonicated total yeast DNA fragments from FY1679 ranging in size from 200 to 1,200 bp were treated with mung-bean nuclease, T4 DNA polymerase and Klenow enzyme following the manufacturers\' protocols. Blunt ends of DNA fragments were ligated to the following adaptor: 5\'-pATCCCGGACGAAGGCC-3\' 3\'-GGCCTGCTTCCGG-5\'. Excess of unligated adaptors and small adaptor-DNA fragments were eliminated by two consecutive purifications using Chroma spin+TE-400 columns (Clontech). Vector predigested with BamHI and filled in with dGTP by the Vent (exo-) polymerase (New England Biolabs) was ligated to the purified adaptor-DNA inserts (800 ng = \~0.16 pmol vector, 800 ng = \~1.7 pmol inserts, in a 40 μl final volume per ligation). The ligation result is drawn in Figure [1b](#F1){ref-type="fig"}. Electroporations of 40 μl of E. coliDH10B cells were performed with 1.8 μl of ligation mix and plated onto 2YT medium (16.1 g/l Bacto tryptone, 10.1 g/l Bacto yeast extract, 5 g/l NaCl, 15 g/l Bacto agar) containing 100 μg/ml ampicillin (four 12 × 12 cm plates per transformation) giving 25,000 to 45,000 clones per transformation. A total of 51 independent transformations were made. This corresponds to 1,888,000 clones. We tested 150 clones for the presence of an insert and observed that more than 85% contained one (average size 700 bp, minimum 220 bp, maximum 1,620 bp). Colonies from each transformation were pooled and distinct Qiagen Tip 500 DNA preparations were made and stored separately for yeast transformation. Final concentration of DNA was 300 to 1,300 ng/μl. The detailed protocol of library construction is available on request. Another library had previously been constructed with the same vector ligated to a distinct DNA-adaptor preparation and was partially used, giving rise to 160,000 primary clones. Characteristics of the transformants were the same as described above. Eight pools of plasmid DNA were prepared from this first library. Yeast transformations --------------------- We carried out a total of 28 independent transformations of yeast by the LiAc method \[[@B55]\]: five with the yeast strain FYAT-01 using five distinct plasmid DNA preparations from the first library and 23 with the strain FYBL2-5D using 23 distinct plasmid DNA preparations from the main library. Aliquots of each transformation were spread onto 24 × 24 cm plates (Q-Pix Trays, Genetix) containing SC - URA + doxycycline, to obtain 1,000 to 3,000 yeast transformants per plate. Screening and storage of toxic clones ------------------------------------- Transformed yeast clones were transferred into fresh liquid SC - URA + doxycycline medium in 96-well microplates by manual picking (30,015 clones) or with the Q-Pix robot (54,071 clones) for overnight growth. Non-toxic and toxic control clones (transformed by empty pCMha190 vector and by vector bearing MCM1, respectively) were also inoculated into each microplate. Cultures were grown overnight at 30°C and stored at 4°C before dilutions for phenotypic examination. Screening of the toxic phenotypes after overexpression was done in a two-round selection, using the \'drop test\', which allowed us to see even slightly impaired growth effects. Ten-fold serial dilutions in water were made from each 96-well culture microplate with a Beckman Biomek 2000 robot, then manually replicated with the 96-pin Beckman replicator onto SC - URA + doxycycline and SC - URA plates in parallel (Figure [1c](#F1){ref-type="fig"}). Clones showing impaired growth in overexpression conditions were streaked onto SC - URA + doxycycline medium for colony isolation, then transferred (one subclone per streak) into a new 96-well microplate and grown for 22 h at 30°C. This plate served as a mother plate for four culture microplates which were grown overnight at 30°C (one plate for the second-round screening, another plate for PCR amplification on colonies for sizing and sequencing the inserts and two plates for storage at -80°C). For the second round of screening, cultures were diluted (1/100 to 1/10,000 dilution) and tested on SC - URA + doxycycline and SC - URA plates in parallel. Phenotypes in the presence and absence of doxycycline (uninduced and overexpression conditions respectively) were scored as described in Results and Figure [2](#F2){ref-type="fig"}. Between the two rounds of screening, most of the clones conserved a comparable phenotype. For those displaying an important difference, a new subclone was tested again, and the transformant was rejected if the phenotype revealed was inconsistent. The dependence of the phenotype on the presence of the plasmid was demonstrated using two methods: for 150 tested clones, wild-type phenotypes were recovered after plasmid loss using 5-FOA resistance selection; for 35 other clones, plasmids were extracted from transformed strain FYAT-01 and retransformed into strain FYBL2-5D, in which the toxic phenotypes were confirmed. Identification of the toxic inserts at the nucleotide and peptide levels ------------------------------------------------------------------------ Inserts of the selected clones were PCR-amplified directly from cultures using primers SEQ4 and SEQ8 (Figure [1b](#F1){ref-type="fig"}). The length of each insert was determined by gel electrophoresis and the 5\' junction was sequenced using primer SEQ1. Identification of each insert in our internal database (see below) was carried out using the DOGEL program \[[@B56]\], adapted by Nicolas Joly (Institut Pasteur) to our purpose. This program gives the start position of the insert on chromosomes, the corresponding genetic object and the start position in the ORF relative to the natural ATG (see Additional data files 1 and 2). We first verified the sequence at the junction with the adaptor-insert. Correct in-frame ligation between vector and adaptor-1 was observed for 632 clones (88.5% of total). For the remaining 82 clones, base substitutions, and short (one to three nucleotides) deletions within the adaptor-1 were observed (nine and 18 cases respectively). A total of 46 cases of a single G addition at the junction vector-adaptor-1, and 15 partial vector sequence duplicates were found (see Additional data files 1 and 2). As the incorrect ligations introduced no stop codon between the initiation codon of the vector and the first codon of the insert, these clones were conserved for further analysis. In these cases, the start position of the insert relative to the chromosome and to the ORF coordinates was corrected manually. For analysis of in-frame ORF fragments, sequences of peptides encoded by toxic inserts were extracted from the complete sequences of S. cerevisiaeproteins, taking the first amino acid corresponding to the junction with the adaptor as the starting point and the end of the insert or the last codon of the ORF as the end point. Fragments of mtDNA, 2 μm plasmid, and DNA coding for Y\'-ORFs, Tys, long terminal repeats (LTRs) and RNA were examined manually for their position relative to the coding sequences. Sequences of inserts other than in-frame ORF fragments were systematically translated into amino-acid sequences from the junction with the adaptor up to the first stop codon encountered in the insert. Sequences coding for more than 24 amino acids were internally compared using BLASTP, then compared to the S. cerevisiaeannotated ORFs and to the 308,738 sequences of our internal database (see below). Databases --------- Genetic entities corresponding to the toxic inserts were identified by comparison with the DNA sequences of the 16 chromosomes (available in the Comprehensive Yeast Genome Database (CYGD) at MIPS \[[@B38]\]); with our own interpretation table containing the coordinates of 6,256 coding sequences (CDS or ORFs), which comprises the new genes found by Blandin et al.\[[@B57]\]; with the 2 μm plasmid DNA sequence \[[@B58]\]; and with the yeast mitochondrial sequence \[[@B59]\]. The set of 6,256 ORFs of S. cerevisiaewas filtered to eliminate all spurious ORFs or unlikely real genes, as well as Ty, Y\' and mitochondrial ORFs, yielding a final list of 5,803 ORFs \[[@B60]\]. For all comparisons of the set of 454 toxic ORFs with the set of ORFs of the entire genome, we used these 5,803 ORFs. GPROTEOME3 is an updated version of the GPROTEOME sequence library \[[@B61]\] containing 308,738 predicted protein sequences from 60 organisms (F. Tekaia, personal communication). Analysis of the toxic inserts and of their cognate genes -------------------------------------------------------- Comparisons among the peptides encoded by in-frame ORF fragments were done using BLASTP \[[@B62]\]. Alignments corresponding to E-values equal to or lower than 10^-3^were examined individually before validation. Conserved domains or patterns of COGs \[[@B32]\] were identified using the NCBI Conserved Domain Search service (CD-Search \[[@B63],[@B64]\]). The NCBI Conserved Domain Database (cdd.v1.62) \[[@B65]\] contained domains derived from Smart \[[@B66]\] and Pfam \[[@B67]\] collections, plus contributions from NCBI such as COGs, leading to 11,088 position-specific score matrices (PSSMs). A routine was written for extraction of the CD-Search results obtained for the toxic inserts and the 5,803 proteins of the entire genome. The cut-off E-value was chosen to be equal to or less than 10^-4^for most domains, and 10^-3^for short domains (60 amino acids or fewer). Domains were considered as present even when represented only partially. In describing genes (Table [4](#T4){ref-type="table"}) or toxic in-frame inserts (see Additional data files 1, 3 and 4), only one domain (giving the best hit) was chosen for a given insert, among several possible hits. In contrast, to compare the frequency of a given domain among all toxic inserts versus its frequency among the 5,803 proteins of S. cerevisiae(Table [2](#T2){ref-type="table"}), all occurrences were taken into account, giving a total of 843 occurrences among the 493 toxic inserts, and a total of 15,925 occurrences among the 5,803 proteins. Searches for transmembrane spans (TMS) were done using TopPredII \[[@B68]\] implemented by Deveaud and Schuerer (Institut Pasteur), predicting both certain and putative TMS. The isoelectric points (IEPs) of proteins or peptides were calculated using iep algorithm from the European Molecular Biology Open Software Suite (EMBOSS) \[[@B69]\]. Descriptions of selected genes and their products were retrieved from the Yeast Proteome Database \[[@B70]\] (release of March 2002; this database is no longer freely available), and from MIPS \[[@B38]\]. Functional classes, cellular localizations and a list of essential genes were retrieved from MIPS \[[@B38]\]; gene classes (conserved/asco-specific/orphan) are from Génolevures \[[@B37]\]. Paralogous gene families of S. cerevisiae\[[@B57]\] are accessible at Génolevures \[[@B37]\] through gene or ORF name. We searched for the participation of the selected ORFs in protein-protein interactions (genetic and physical) and in protein complexes using three different sources: YPD \[[@B70]\] files for individual proteins; protein complexes defined by Gavin et al.\[[@B10]\]; data compilations concerning protein-protein interactions and complexes, extracted from SGD \[[@B1]\], MIPS \[[@B38]\] and unpublished two-hybrid experiments (M. Fromont-Racine and C. Saveanu, personal communication). How representative is our screening? ------------------------------------ We consider that our library contains DNA fragments randomly distributed throughout the genome. Out of 84,086 clones tested, 11% (9,530) contain a DNA fragment cloned in-frame with the frame of the natural ORF (\~68% of the genome corresponds to coding regions, and only one frame out of six corresponds to the natural frame), the others containing noncoding, out-of-frame or antisense DNA fragments. If we use the simplifying assumption that all genes are equally represented among the 9,530 clones (not taking into account the size diversity of genes), each of the 5,803 ORFs will be represented 1.64 times (9,530/5,803). The probability P~x~of encountering any gene xtimes is described by a Poisson distribution: ![](gb-2004-5-9-r72-i1.gif) where m, the mean of the distribution, is 1.64. This is used to estimate the fraction of genes not encountered: for x= 0 and probability p= 0.19, the number of non-encountered genes = 1,126. Thus, by screening a total of 84,086 clones, we have encountered a maximum of 4,677 ORFs (5,803 - 1,126). Additional data files ===================== The following additional data are available with the online version of this article. Additional data file [1](#s1){ref-type="supplementary-material"} contains lists and coordinates of the 493 in-frame fragments of annotated ORFs giving toxic phenotypes when overexpressed, and short description of their cognate genes. Additional data file [2](#s2){ref-type="supplementary-material"} contains a list and description of the 221 DNA toxic inserts other than in-frame ORF fragments. Additional data file [3](#s3){ref-type="supplementary-material"} gives a description of the peptides encoded by the 493 toxic ORF fragments, and of the cognate proteins. Additional data file [4](#s4){ref-type="supplementary-material"} gives the content of the 57 groups of peptide inserts sharing similarities. Additional data file [5](#s5){ref-type="supplementary-material"} gives a list and description of protein domains found only once among the toxic inserts. Additional data file [6](#s6){ref-type="supplementary-material"} lists the genes selected in this work whose products are members of complexes \[[@B10]\]. Additional data file [7](#s7){ref-type="supplementary-material"} lists genes selected in this work whose products are known as interacting with each other. Additional data file [8](#s8){ref-type="supplementary-material"} contains the sequences of the oligonucleotides used in this work. Additional data file [9](#s9){ref-type="supplementary-material"} contains a figure showing the phenotypes induced by overexpression of antiparallel ORF fragments before and after introduction of a stop codon upstream of the artificial ORFs. Supplementary Material ====================== ::: {.caption} ###### Additional data file 1 Lists and coordinates of the 493 in-frame fragments of annotated ORFs giving toxic phenotypes when overexpressed, and short description of their cognate genes ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 2 A list and description of the 221 DNA toxic inserts other than in-frame ORF fragments ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 3 A description of the peptides encoded by the 493 toxic ORF fragments, and of the cognate proteins ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 4 The content of the 57 groups of peptide inserts sharing similarities ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 5 A list and description of protein domains found only once among the toxic inserts ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 6 The genes selected in this work whose products are members of complexes ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 7 Genes selected in this work whose products are known as interacting with each other ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 8 The sequences of the oligonucleotides used in this work ::: ::: {.caption} ###### Click here for additional data file ::: ::: {.caption} ###### Additional data file 9 A figure showing the phenotypes induced by overexpression of antiparallel ORF fragments before and after introduction of a stop codon upstream of the artificial ORFs ::: ::: {.caption} ###### Click here for additional data file ::: Acknowledgements ================ We thank F. Tekaia and N. Joly for suggestions and support during this work, E. Couvé for help with experiments, M. Fromont-Racine and C. Saveanu for helpful discussions and communications of unpublished data, and S. Marchiset, C. Lequatre and L. Oreus for media supply and technical help. This work was supported in part by the EUROFAN2 project (Bio4-CT97-2294) from the European Commission (DGXII). E.T. was supported by the European contract CYGD (QLRI-CT 1999-01333), R.K. is a recipient of a CNRS-BDI fellowship. B.D. is a member of the Institut Universitaire de France. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Overexpression library construction and screening. (a)Construction of an HA-tagged vector. The pCMha190 vector used here was constructed by insertion of a linker (gray box) in place of the multiple cloning site in vector pCM190 \[31\]. Features shown include the promoter and TATA box as well as the terminator from the original plasmid (open boxes), and the start codon, HA-tag, BamHI site and stop codons (thick vertical bars) from the introduced linker sequence. The linker was composed from the following annealed oligonucleotides: EXP3: 5\'-GATCGTTTAAACCATATGTACCCATACGACGTCCCAGACTACGCTGG ATCCTGACTGACTGATC-3\', EXP4: 5\'-GGCCGATCAGTCAGTCAGGATCCAGCGT AGTCTGGGACGTCGTATGGGTACATATGGTTTAAAC-3\'. (b)Library construction in pCMha190 (see Materials and methods for experimental details). The resulting ligation product is schematized, with the insert as a striped box and adaptors as hatched boxes. Sequences shown below are from junctions, with uppercase letters corresponding to vector (the extra nucleotide from filling-in is underlined), lowercase letters to adaptors and bold nnn\'s to insert. Arrows indicate the different primers used: SEQ8 and SEQ4 are used for PCR amplification of the insert, and SEQ1 for sequencing (see sequences in Additional data file 8). (c)First-round screening of toxic phenotypes. The growth of random and control clones on selective medium in uninduced and overexpression conditions is shown. Drops of serial dilutions (1/100 to 1/100,000) of cultures were grown for 45 h at 30°C. A3, non-toxic control clone transformed by pCMha190; H1, toxic control clone transformed by MCM1gene cloned in pCMha190; G1, B2, D2, E3, library transformed clones, exhibiting different levels of toxicity in overexpression conditions (see Figure 2). ::: ![](gb-2004-5-9-r72-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Second-round scoring of toxic phenotypes and control. (a)Selected clones from the first round were diluted and three drops (1/100, 1/1,000 and 1/10,000) were spotted and grown for 42 h at 30°C, with controls on same plates, for confirmation of toxicity. Growth levels in the presence and absence of doxycycline were scored as described in the text. Each clone was assigned a growth index where the first number represents the growth in uninduced conditions and second number the growth in induced conditions; for example, 3/3 indicates a non-toxic insert; 3/0 indicates a highly toxic insert. Clone numbers are the same as in the tables describing the toxic inserts (see Additional file 1,2,3,4). (b)After 5-FOA-induced plasmid loss, growth of surviving clones is scored in the same way as in (a). Wild-type phenotypes in overexpression conditions are indicative of plasmid-borne toxicity. ::: ![](gb-2004-5-9-r72-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Toxic phenotypes of overexpressed fragments versus whole ORF products. Complete ORFs are cloned in pCMha191 (tryptophan marker); inserts are cloned in pCMha190 (uracil marker). Eleven out of the 13 cases are represented in this figure. + doxycycline, uninduced conditions; - doxycycline, overexpressed conditions. ::: ![](gb-2004-5-9-r72-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Positions of selected toxic fragments relative to the structure of genes of the PI kinase family. Names of the selected genes and protein lengths (in amino acids) are indicated. Coordinates of the toxic fragments selected in this work and of known toxic domains (see text) are also given. Conserved domains in the proteins have been positioned using the NCBI CD-Search program \[64\] (see Materials and methods). Domain abbreviations: FAT (pfam 00259) is named after FRAP, ATM and TRRAP, which are human homologs of yeast TOR, TEL1 and TRA1, respectively; PI3Kc (smart00146) is the PI kinase catalytic domain; FATC (pfam02260.11) is named after FRAP, ATM, TRRAP carboxy-terminal region. Complete COG5032 TEL1 (2,105 residues) spans the carboxy-terminal regions of the four proteins. The drawing is not to scale. ::: ![](gb-2004-5-9-r72-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Distribution of the toxic inserts between the different genetic objects ::: Genetic objects represented Number of toxic inserts Percentage of total Mean size ± SD (nucleotides) (minimum-maximum) Phenotypes Inserts encoding artificial peptides ----------------------------- ------------------------- --------------------- ------------------------------------------------ ------------ -------------------------------------- ---- ---- ----- In-frame ORF fragments 493 68.7 743 ± 311 (220-2,120) 375 87 23 8 \_ Antiparallel ORF fragments 68 9.6 532 ± 247 (140-1,220) 37 11 12 8 53 Out-of-frame ORF fragments 53 7.5 733 ± 306 (170-1,620) 12 11 22 8 12 Intergenic regions 41 6.0 625 ± 358 (170-1,820) 13 4 16 8 27 LTRs 2 0.3 595 (320-1,120) 1 0 0 1 1 Ty elements 15 (10) 2.1 633 ± 265 (320-870) 7 4 2 2 \_ Y\' elements 9 (3) 1.2 678 ± 370 (320-1,320) 9 0 0 0 6 RNA genes 4 0.5 662 ± 246 (470-1,020) 3 0 1 0 3 2 μm plasmid 17 (10) 2.4 564 ± 288 (170-1,220) 13 3 1 0 5 Mitochondrial DNA 12 1.7 483 ± 201 (200-920) 9 3 0 0 10 Total 714 100 703 ± 313 (140-2,120) 479 123 77 35 117 The first column indicates nature of sequence in toxic inserts. Second and third columns contain, respectively, actual number of inserts of each type and corresponding percentages. For Tys, Y\' and 2 μm plasmid, numbers in brackets represent numbers of in-frame fragments of natural ORFs. The fourth column shows the mean size of insert in nucleotides ± standard deviation (SD) with minimum and maximum sizes in brackets. Scoring of each type of phenotype is shown in the next four columns. The last column shows the number of inserts in which artificial ORFs of more than 24 codons were detected. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Conserved domains found more than once among the toxic in-frame ORF fragments ::: Domain reference Domain name S. cerevisiae Toxic inserts Mean 95% confidence interval Result Domain description ------------------------------------------------------------------------------- ---------------------- ----------------- --------------- ------- ------------------------- -------- -------------------------------------------------------------------------------- Transport-specific domains COG0471 CitT 4 4 0.21 0.17-1.25 \+ Di-and tricarboxylate transporter pfam03169 OPT 3 3 0.16 0.11-1.17 \+ Oligopeptide transporter protein COG1953 FUI1 9 3 0.48 0.44-1.56 \+ Nucleotide transporter pfam00324 aa\_permeases 22 7 1.16 1.04-2.22 \+ Amino acid permease pfam00153 mito\_carr 97 24 5.13 5.07-6.45 \+ Mitochondrial carrier protein COG0531 PotE 26 5 1.38 1.28-2.48 \+ Amino acid transporter COG0474 MgtA 23 4 1.22 1.12-2.30 \+ Cation transport ATPase cd00267 ABC\_ATPase 58 6 3.07 2.93-4.22 \+ ABC transporter nucleotide-binding domain pfam00664 ABC\_membrane 14 2 0.74 0.68-1.82 \+ ABC transporter transmembrane region COG0842 COG0842 6 3 0.32 0.29-1.38 \+ ABC-type multidrug transport system, permease component COG1131 CcmA 54 4 2.86 2.74-4.01 NS ABC-type multidrug transport system, ATPase component pfam00083 Sugar\_tr 58 5 3.07 2.94-4.23 \+ Sugar (and other) transporter RNA-and DNA-binding domains pfam00076 rrm 72 11 3.81 3.62-4.95 \+ RNA recognition motif (transcription) COG5099 (PUF) 9 5 0.48 0.44-1.56 \+ Pumilio family RNA-binding repeat (translational repression) smart00322 KH 11 4 0.58 0.54-1.66 \+ K homology: RNA-binding domain (transcription, RNA metabolism) smart00356 ZnF\_C3H1 5 4 0.26 0.21-1.30 \+ Zinc finger, C3H1 type (transcription) COG5048 C2H2-type Zn\_finger 15 4 0.79 0.74-1.89 \+ Zn-finger (C2H2-type) (transcription) COG0210 UvrD 4 2 0.21 0.17-1.24 \+ DNA and RNA helicases, superfamily I (DNA replication, recombination, repair) cd00086 Homeodomain 9 2 0.48 0.45-1.57 \+ DNA binding domain (eukaryotic development) pfam00249 myb\_DNA-binding 13 2 0.69 0.66-1.80 \+ Myb-like DNA-binding domain (transcription) pfam00170 bZIP 4 2 0.21 0.17-1.25 \+ Basic-leucine zipper DNA binding and dimerization domains (transcription) smart00066 GAL4 48 2 2.54 2.44-3.72 NS GAL4-like Zn(II)2Cys6 DNA-binding domain (fungal) (transcription) pfam04082 Fungal\_trans 26 2 1.38 1.29-2.48 NS Fungal specific transcription factor domain. pfam00270 DEAD 48 3 2.54 2.38-3.63 NS DEAD/DEAH box helicase (replication, repair, transcription) cd00079 HELICc 60 2 3.18 3.08-4.34 \_ Helicase superfamily, C-ter domain (replication, repair, transcription) Domains involved in Interactions with peptides, proteins or phospholipids cd00200 WD40 327 29 17.31 16.87-18.54 \+ Tandem repeats of about 40 residues interacting with peptides pfam01602 Adaptin\_N 9 2 0.48 0.43-1.54 \+ N-ter region of adaptor proteins (clathrin-coated pits and vesicles) pfam00786 PBD 4 2 0.21 0.20-1.27 \+ P21-Rho-binding domain (or CRIB) pfam00169 PH 11 3 0.58 0.55-1.67 \+ PH: pleckstrin homology. binds phosphoinositides or other ligands (signalling) COG5271 MDN1 16 3 0.85 0.78-1.93 \+ AAA : ATPase with von Willebrand factor type A domain (multiprot. complexes) smart00268 ACTIN 14 2 0.74 0.67-1.82 \+ ACTIN, cytoskeleton/motor protein COG5022 Myosin heavy chain 7 5 0.37 0.33-1.43 \+ ATPase, molecular motor COG5043 MRS6 4 2 0.21 0.17-1.24 \+ Vacuolar protein sorting-associated protein KOG0446\* Dynamin 3 3 0.16 0.13-1.20 \+ GTPase that mediates vesicle trafficking Metabolism-related domains pfam03901 PMP 5 2 0.21 0.21-1.29 \+ Mannosyltransferase COG1928 PMT1 7 4 0.37 0.30-1.40 \+ Mannosyltransferase pfam00561 Abhydrolase 18 3 0.95 0.88-2.05 \+ Abhydrolase, alpha/beta hydrolase fold (catalytic domain) pfam00107 ADH\_zinc\_N 21 2 1.11 1.01-2.19 NS Zinc-binding dehydrogenase pfam00501 AMP-binding 11 2 0.58 0.51-1.64 \+ AMP-binding synthetase Other domains pfam00674 DUP 35 3 1.85 1.81-3.03 NS DUP family (proteins of unknown functions) COG5384 Mpp10 1 2 0.05 0.03-1.07 \+ M phase phosphoprotein 10 (U3 small nucleolar ribonucleoprotein component) COG5032 TEL1 8 4 0.42 0.34-1.44 \+ PI kinase and protein kinases of the PI kinase family COG1025 Ptr 5 2 0.26 0.22-1.31 \+ Zn-dependent peptidases (secreted/periplasmic, insulinase-like) pfam02902 Peptidase\_C48 2 2 0.11 0.08-1.13 \+ Ulp1 protease family, C-terminal catalytic domain pfam00004 AAA 43 3 2.28 2.15-3.39 NS AAA, ATPase family associated with various cellular activities (AAA) smart00220 S\_TKc 125 4 6.52 6.31-7.72 \- Serine/threonine protein kinases, catalytic domain Peptide sequences of toxic natural ORF fragments were searched for domains (see text), and the frequency of domains found more than once was compared to the frequency in the whole proteome. References and names of domains are in the first two columns; occurrences in the whole genome (S. cerevisiae) and in the toxic inserts are in the third and fourth columns, respectively. The next three columns show the statistical analysis performed as follows: 1,000 random selections of 843 domains (total number of occurrences in the toxic inserts) were made from the set of 15,925 domains identified in S. cerevisiae(see Materials and methods); mean (column 5) represents the mean number of occurrences of each domain among the toxic inserts; the 95% confidence interval (column 6) was calculated using the SD of the 1,000 random drawings; column 7 shows the result of this analysis for each domain: NS, not significant; +, domain over-represented in toxic inserts; -, domain under-represented in toxic inserts. The last column gives a brief description of domains from NCBI Conserved Domain Database \[65\]. \KOG0446 was found using cdd.v1.63 of NCBI CD-Search \[64\]. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Distribution of selected genes versus all S. cerevisiaegenes ::: All S. cerevisiaegenes Percentage of total Selected toxic genes Percentage of total ------------------------------------------------------ -------------------------- --------------------- ---------------------- --------------------- Functional classes (MIPS data)* Cell cycle\_DNA processing 670 11.5 75 16.5\* Cell fate 486 8.4 66 14.5\* Cell rescue, defense and virulence 288 5.0 23 5.1 Cellular communication/signal transduction mechanism 59 1.0 6 1.3 Cellular transport and transport mechanisms 525 9.0 67 14.8\* Classification not yet clear-cut 112 1.9 6 1.3 Control of cellular organization 207 3.6 22 4.8 Energy 244 4.2 12 2.6 Metabolism 1,061 18.3 88 19.4 Protein fate (folding, modification, destination) 593 10.2 47 10.4 Protein synthesis 377 6.5 17 3.7\* Regulation of/interaction with cell. Environment 197 3.4 29 6.4^†^ Transcription 801 13.8 88 19.4\* Transport facilitation 321 5.5 61 13.4\* Unclassified 1,706 29.4 91 20.0\* Cellular localization (MIPS data) Extracellular 54 1.4 5 1.6 Cell wall 38 1.0 4 1.3 Golgi 103 2.6 8 2.5 Transport vesicles 54 1.4 3 0.9 Plasma membrane 171 4.4 34 10.7\* Nucleus 1,367 34.8 130 40.8^†^ Cytoplasm 2,001 50.9 137 42.9\* Peroxisome 42 1.1 3 0.9 Endosome 20 0.5 2 0.6 Cytoskeleton 154 3.9 22 6.9^†^ Vacuole 82 2.1 8 2.5 Endoplasmic reticulum 353 9.0 27 8.5 Mitochondria 562 14.3 37 11.6 Viability (MIPS data) Essential 939 16.2 96 21.1^†^ Essential or not 160 2.8 20 4.4 Phylogeny (Génolevures data) Conserved 3,717 64.1 336 74.0\* Ascomycete-specifics 1674 28.8 106 23.3\* Orphan 412 7.1 10 2.2\* The distribution of genes was examined in respect of four classifications: function, cellular localization of the gene product, viability and phylogeny. Data are from MIPS \[38\] and Génolevures \[37\]. Cellular localization was known for 3,928 out of the 5,803 proteins in the entire genome and for 319 proteins out of the 454 that yield toxic inserts. For other comparisons, the set of 454 selected genes was compared to the set of 5,803 genes of S. cerevisiae. Note that a given gene may be present in more than one MIPS class. Significant evidence that a given gene class is over-or under-represented among toxic genes as compared to all S. cerevisiaegenes is emphasized by bold characters. \*p\< 0.005; ^†^p\< 0.025. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Toxicity of fragments versus whole ORF products ::: ORF/Gene name Gene description Phenotype of gene deletion Conserved domain or TMS in entire protein Phenotype of gene overexpression Conserved domain or TMS in insert Phenotype of insert overexpression ------------------------- -------------------------------------------------------------- ---------------------------- ------------------------------------------- ---------------------------------- ----------------------------------- ------------------------------------ YDL112w/TRM3\* tRNA 2\'-O-ribose methyltransferase Viable SpoU\_methylase 3/3 \- 3/1 YML128C/MSC1/ GIN3\^†^ Weak similarity to Schizosaccharomyces pombestress protein Viable 1 TMS 3/3 \- 3/0 YGR149w/\_\ ^‡§^ Similar to S. pombehypothetical protein Viable 5 TMS 3/2 to 3/3 3 TMS 3/2 YGL023c/PIB2\* ^§^ Phosphatidylinositol 3-phosphate binding Viable FYVE 3/1 FYVE 3/0 YPL043w/NOP4^¶^ Nucleolar protein, RNA processing Lethal RRM (4 motifs) 3/0 Bias D, E, K 3/0 YOR166c/\_ \* ^§^ Similarity to hypothetical S. pombeprotein Viable PINc (nucleotide binding) 3/0 PINc 3/0 YJL212c/OPT1^¶¥^ Oligopeptide transporter Viable OPT 3/1 2 TMS, OPT 3/1 YNL003c/PET8^¥¤^ Mitochondrial carrier Viable mito\_carrier 3/2 mito\_carrier 3/2 YJL092w/HPR5^¥\#^ DNA helicase involved in DNA repair Viable UvrD 2/0 UvrD (central) 3/2 YMR190c/SGS1^¥¤^ DNA helicase of DEAD/DEAH family Viable DEAD, HELICc, HRDC 3/0 DEAD 3/2 YNL033W/\_^§^ Strong similarity to YNL019c Viable 2 TMS 3/1 1 TMS 3/2 YHR067w/\_\* ^§^ Weak similarity to S. pombehypothetical protein Viable Maoc : Acyldehydratase 3/1 MaoC 3/2 YGL263w/COS12^§¥¤^ Similarity to subtelomeric encoded proteins Viable DUP 3/0 DUP 3/1 Systematic nomenclature and gene name, where applicable, are given in the first column. \Singleton: the gene has no paralog in S. cerevisiae. ^†^Gene fragment and ^\#^entire gene, respectively, were already known as toxic upon overexpression. ^‡^Putative uncharacterized transporter (see \[35\]). ^§^Gene of unknown classification. ^¶^Two non-overlapping inserts of the ORF were selected. ^¥^One or several paralogs of this gene have also been selected as toxic inserts in this work (see Additional data file 3). ^¤^Gene having a paralog in S. cerevisiae*already known as toxic upon overexpression. Columns 2 and 3 contain respectively a brief description of the function of the gene product and the phenotype of the disruption mutant (MIPS \[38\]). The results of a search for conserved domains is shown in columns 4 (in whole protein) and 6 (in inserts). Phenotypes in uninduced and overexpression conditions of the entire gene and of fragments are given in columns 5 and 7 respectively (see Figure 3 for illustrations of the phenotypes). :::	PubMed Central	2024-06-05T03:55:48.543778	2004-8-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522879/", "journal": "Genome Biol. 2004 Aug 31; 5(9):R72", "authors": [ { "first": "Jeanne", "last": "Boyer" }, { "first": "Gwenaël", "last": "Badis" }, { "first": "Cécile", "last": "Fairhead" }, { "first": "Emmanuel", "last": "Talla" }, { "first": "Florence", "last": "Hantraye" }, { "first": "Emmanuelle", "last": "Fabre" }, { "first": "Gilles", "last": "Fischer" }, { "first": "Christophe", "last": "Hennequin" }, { "first": "Romain", "last": "Koszul" }, { "first": "Ingrid", "last": "Lafontaine" }, { "first": "Odile", "last": "Ozier-Kalogeropoulos" }, { "first": "Miria", "last": "Ricchetti" }, { "first": "Guy-Franck", "last": "Richard" }, { "first": "Agnès", "last": "Thierry" }, { "first": "Bernard", "last": "Dujon" } ] }
PMC523228	Introduction {#s1} ============ Formation of organs depends on successive programs of gene expression during development. Temporal regulation of transcription is critical to achieve spatial patterning, as seen during vertebrate somitogenesis ([@pbio-0020352-Pourquie1]). Temporal regulation is also essential to integrate the progression of events that accompany cell fate specification and differentiation ([@pbio-0020352-Bruhn1]; [@pbio-0020352-Isshiki1]; [@pbio-0020352-Pearson1]). The regulatory networks that guide these processes depend on a wide array of transcription factors, raising the question of how transcriptional circuitry dictates developmental timing. In some cases, tiers of transcription factors function hierarchically to establish sequential patterns of gene expression ([@pbio-0020352-Kornberg1]; [@pbio-0020352-Maduro1]; [@pbio-0020352-Skeath1]). These regulators are active for only a brief time during development and typically produce a uniform response in expressing cells. For example, three consecutive waves of GATA transcription factors establish the Caenorhabditis elegans midgut ([@pbio-0020352-Maduro1]). Ectopic expression of at least some of these GATA factors can convert the entire embryo into midgut, suggesting a homogeneous and robust transcriptional response by these cells. In contrast, other transcriptional regulators function continuously during development ([@pbio-0020352-Weatherbee1]; [@pbio-0020352-Bergstrom1]; [@pbio-0020352-Gaudet1]). These proteins activate different target genes at different developmental stages, suggesting a more complex, heterogeneous transcriptional response. A critical question is how the second class of transcriptional regulators establishes consecutive programs of gene expression. The forkhead box (Fox) A family of transcription factors illustrates the second strategy of developmental control. These proteins are critical to form the digestive tract in Drosophila, mammals, and worms, and animals lacking FoxA have profound gut defects ([@pbio-0020352-Weigel1]; [@pbio-0020352-Ang1]; [@pbio-0020352-Mango1]; [@pbio-0020352-Weinstein1]; [@pbio-0020352-Dufort1]). For example, inactivation of C. elegans pha-4 leads to a loss of foregut cells, which are transformed into ectodermal cell types such as glia and epidermis ([@pbio-0020352-Mango1]; [@pbio-0020352-Horner1]). This dramatic phenotype reflects the global requirement for PHA-4 to transcribe genes selectively expressed in foregut cells throughout development. Direct PHA-4 targets include early-acting developmental regulators, such as ceh-22/Nkx 2--5, and terminal differentiation genes that encode structural proteins and digestive enzymes ([@pbio-0020352-Kalb1]; [@pbio-0020352-Gaudet1]; [@pbio-0020352-Vilimas1]). FoxA members in other organisms have a similar range of early- and late-expressed targets, suggesting they, too, function during multiple stages of development ([@pbio-0020352-Gualdi1]; [@pbio-0020352-Lehmann1]; [@pbio-0020352-Duncan1]; [@pbio-0020352-Roth1]; [@pbio-0020352-Lee1]). The diversity of FoxA target genes raises the question of how these factors achieve appropriate temporal regulation of transcription during development. One answer is affinity for DNA. C. elegans PHA-4 recognizes sequences that conform to the consensus TRTTKRY (where R = A/G, K = T/G, and Y = T/C) ([@pbio-0020352-Overdier1]; [@pbio-0020352-Gaudet1]). Sequences that bind PHA-4 with high affinity in vitro are typically found in promoters of genes expressed early in development, whereas low-affinity sites are restricted to late promoters ([@pbio-0020352-Gaudet1]). Moreover, adjustment of a high-affinity binding site to a lower one shifts the onset of expression later, and, conversely, mutation to a higher-affinity site leads to earlier activation ([@pbio-0020352-Gaudet1]). These data demonstrate that binding-site affinity of PHA-4 for DNA is a critical determinant of gene expression. However, the affinity of PHA-4 for its recognition sequence is not an absolute predictor of gene activation. For example, the pharyngeal muscle myosin gene, myo-2, possesses high-affinity PHA-4 sites but is activated late in development ([@pbio-0020352-Okkema2]; [@pbio-0020352-Gaudet1]). These observations suggest additional factors function in combination with PHA-4 for temporal control of pharyngeal gene expression. In this study, we have combined bioinformatics and experimental approaches to investigate the cis-regulatory network for temporal gene expression within the pharynx. We identify sites that function in combination with PHA-4 elements to distinguish early from late expression. These elements can be used to build synthetic promoters with the expected expression profiles and to identify previously undiscovered pharyngeal genes within the genome. Results {#s2} ======= Our goal was to discover new regulators of pharyngeal transcription that would function in combination with PHA-4 ([Figure 1](#pbio-0020352-g001){ref-type="fig"}A). To achieve this aim, we first identified candidate pharyngeal genes by microarray analysis and subdivided these genes into clusters based on their onset of expression (early versus late). Next, we searched for short sequences enriched within the predicted promoters of genes from the early or late clusters. These sequences were tested for enhancer or repressor activity in vivo using both natural and synthetic promoters. ::: {#pbio-0020352-g001 .fig} Figure 1 ::: {.caption} ###### Strategy to Identify Temporal Regulatory Elements \(A) Flowchart of the strategy used. \(B) Northern blot of pha-4. pha-4 transcripts were approximately 25- to 100-fold enriched in par-1 compared to skn-1 embryos, but only approximately 5- to 10-fold enriched in wild-type compared to skn-1 embryos. Arrowheads indicate the three different pha-4 isoforms. \(C) The same blot was probed with a fragment of the act-1 gene to demonstrate equal loading of RNA between lanes. ::: ![](pbio.0020352.g001) ::: Identification of Pharyngeal Genes {#s2a} ---------------------------------- We previously discovered pharyngeal genes using microarrays encompassing 62% of C. elegans genes ([@pbio-0020352-Gaudet1]). We extended this analysis by screening microarrays that covered 94% of C. elegans genes (17, 871 genes; [@pbio-0020352-Jiang1]). To maximize the sensitivity of detection, we compared gene expression profiles from embryos with excess pharyngeal cells (par-1; [@pbio-0020352-Kemphues1]) to embryos with no pharyngeal cells (skn-1; [@pbio-0020352-Bowerman1]). par-1 mutants affect the earliest embryonic cell divisions and produce cell fate transformations, such that par-1 mutant embryos lack gut cells, but have excess pharynx and body wall muscles. In contrast, skn-1 mutants lack both gut and pharynx, but have excess body wall muscle and epidermis. Thus, genes with a relatively high par-1/skn-1 ratio were likely to be selectively expressed in the pharynx. An advantage to using par-1 and skn-1 mutants was that they provided a broad range of expression differences. For example, pha-4 transcripts were approximately 25- to 100-fold enriched in par-1 versus skn-1 embryos, compared to only approximately 5- to 10-fold enriched in wild-type versus skn-1 embryos ([Figure 1](#pbio-0020352-g001){ref-type="fig"}B and [1](#pbio-0020352-g001){ref-type="fig"}C). We identified 339 genes with at least 2-fold greater expression in par-1 embryos compared to skn-1 mutants ([Materials and Methods](#s4){ref-type="sec"}; [Table S1](#st001){ref-type="supplementary-material"}). For genes whose expression was known, 81% (114/141) were selectively expressed in the pharynx ([Table S1](#st001){ref-type="supplementary-material"}). Importantly, the sensitivity of this approach enabled us to detect genes expressed at low levels (e.g., ceh-22; [@pbio-0020352-Okkema1]; [@pbio-0020352-Vilimas1]) or in a small subset of pharyngeal cells (e.g., C49G7.4; [@pbio-0020352-Ao1]). Temporal Groups of Pharyngeal Genes {#s2b} ----------------------------------- We categorized the pharyngeal genes into early versus late temporal groups using previously defined expression patterns. The Nematode Expression Pattern Database ([@pbio-0020352-Kohara1]; <http://nematode.lab.nig.ac.jp/db/index.html>) and green fluorescent protein (GFP) reporters enabled us to identify 37 early-onset and 34 late-onset pharyngeal genes ([Table S2](#st002){ref-type="supplementary-material"}). The set of early-expressed pharyngeal genes (Ph-E) contained genes whose expression initiated by mid-embryogenesis (the "bean" to "comma" stages; [@pbio-0020352-Sulston2]). At the end of this developmental stage, embryonic cell division is virtually complete, the pharynx primordium has formed, and cell fate patterning of the primordium has begun. The set of late-expressed pharyngeal genes (Ph-L) contained genes activated at the onset of terminal differentiation of the pharynx (the 3-fold stage; [@pbio-0020352-Sulston2]). Forty-three genes were associated with expression patterns but were not assigned to either category either because the onset of expression was ambiguous or because it fell between the early and late categories. The Ph-E cluster was enriched for genes predicted to encode transcription factors ([Figure 2](#pbio-0020352-g002){ref-type="fig"}; [Table 1](#pbio-0020352-t001){ref-type="table"}), consistent with Ph-E genes controlling early aspects of pharyngeal development such as cell fate specification. The Ph-L group was enriched for genes predicted to encode cytoskeletal or muscle proteins ([Figure 2](#pbio-0020352-g002){ref-type="fig"}; [Table 1](#pbio-0020352-t001){ref-type="table"}), consistent with Ph-L genes being involved in terminal differentiation and pharyngeal function. Intriguingly, Ph-L genes were more likely to be located on Chromosomes V and X (p ≤ 0.05) at the expense of Chromosomes I and IV (p ≤ 0.06; [Table 2](#pbio-0020352-t002){ref-type="table"}). Biases for gene placement on chromosomes have been observed previously. For example, genes expressed in the male germ line are excluded from the X chromosome ([@pbio-0020352-Reinke1]), while muscle genes are often clustered along a chromosome ([@pbio-0020352-Roy1]). ::: {#pbio-0020352-g002 .fig} Figure 2 ::: {.caption} ###### Different Predicted Products of the Temporal Groups The Ph-E group is enriched for predicted transcription factors, while the Ph-L group is enriched for predicted structural or muscle proteins (see [Table 1](#pbio-0020352-t001){ref-type="table"}). Muscle proteins are proteins known or predicted to be involved in muscle function, including myosins, tropomyosins, and troponins. [Table S2](#st002){ref-type="supplementary-material"} provides a complete listing of the catergorization of the Ph-E and Ph-L genes. ::: ![](pbio.0020352.g002) ::: ::: {#pbio-0020352-t001 .table-wrap} Table 1 ::: {.caption} ###### Temporal Groups of Pharyngeal Genes Are Enriched for Different Kinds of Predicted Products ::: ![](pbio.0020352.t001) Relative to both the genome and the total set of microarray positives, the Ph-E group is enriched for predicted transcription factors. In contrast, the Ph-L group is enriched for predicted structural or muscle proteins. Muscle proteins are proteins known or predicted to be involved in muscle function, including myosins, tropomyosins, and troponins ND, not determined ::: ::: {#pbio-0020352-t002 .table-wrap} Table 2 ::: {.caption} ###### Ph-L Genes Are Enriched on Chromosomes V and X ::: ![](pbio.0020352.t002) ^a^ Expected number if Ph-L genes are randomly distributed throughout the genome ::: Identifying Regulatory Elements in Pharyngeal Promoters {#s2c} ------------------------------------------------------- We examined predicted promoters of Ph-E and Ph-L members for candidate cis-regulatory elements that might contribute to temporal regulation. We first estimated the size of pharyngeal promoters by determining the sequence identity between pairs of C. elegans and C. briggsae orthologs. Conservation is a good indicator of functionally important regions for cis-regulation ([@pbio-0020352-Kirouac1]; [@pbio-0020352-Hwang1]; [@pbio-0020352-Liu1]). We scored as "conserved" those regions of DNA with 75% or greater identity over 50 bp. Sixty-six percent of pharyngeal genes had conserved sequences within 500 bp upstream of the predicted start codon (n = 64 genes; [Table 3](#pbio-0020352-t003){ref-type="table"}), whereas only 21% had conserved sequences from 500 to 1,000 bp (n = 63 genes; [Table 3](#pbio-0020352-t003){ref-type="table"}). This landscape of sequence conservation agreed well with reporter studies in which 500 bp of upstream sequence was often sufficient to recapitulate the endogenous pattern of expression ([@pbio-0020352-Gaudet2]; [@pbio-0020352-McGhee1]; [@pbio-0020352-Gaudet1]). Based on these observations, we chose to limit our motif searches to 500 bp upstream of predicted start codons. ::: {#pbio-0020352-t003 .table-wrap} Table 3 ::: {.caption} ###### Conservation of Non-Coding Sequence between C. elegans and C. briggsae ::: ![](pbio.0020352.t003) Entries show the fraction of genes that have significant conservation of non-coding sequence (greater than 75% identity across 50 bp or more). "−1,000 to −501" is the region 501 to 1,000 bp upstream of the predicted ATG start codon; "−500 to −1" is the region within the first 500 bp upstream of the predicted ATG start codon. For a complete listing of conservation for each gene in the Ph-E and Ph-L groups, see Table S3 ::: We used the Improbizer expectation maximization algorithm to search pharyngeal promoters for potential regulatory elements ([@pbio-0020352-Ao1]). Improbizer detects short sequences that are over-represented within a cohort of genes. As a negative control, we examined groups of genes that were not specifically expressed in the pharynx (e.g., DNA synthesis genes or a set of 194 randomly selected genes) and removed motifs common to both the pharyngeal and negative control sets. This comparison identified nine candidate regulatory motifs for pharyngeal genes ([Figures 3](#pbio-0020352-g003){ref-type="fig"}; [Dataset S1](#sd001){ref-type="supplementary-material"}). Significantly, two motifs, Early-3 and P-3, conformed to the consensus sequences for the pharyngeal transcription factors CEH-22 and PHA-4, respectively ([@pbio-0020352-Okkema1]; [@pbio-0020352-Kalb1]; [@pbio-0020352-Gaudet1]), indicating that our approach could successfully identify pharyngeal regulatory elements. ::: {#pbio-0020352-g003 .fig} Figure 3 ::: {.caption} ###### Candidate Pharyngeal Motifs Identified by Improbizer Improbizer represents motifs as PWMs; the PWMs for the motifs are shown in [Dataset S1](#sd001){ref-type="supplementary-material"}. We converted these matrices to the "sequence logos" shown here ([@pbio-0020352-Schneider1]; [@pbio-0020352-Crooks1]). The threshold scores used by Cluster-Buster ([@pbio-0020352-Frith1]) for each motif are shown below the sequence logos. "% Ph-E Genes" and "% Ph-L Genes" are the percentage of Ph-E and Ph-L genes that contain occurrences of a given motif within 500 bp upstream of the predicted ATG start codon, above the threshold shown. "% Neg Genes" is the percentage of DNA synthesis genes that contain occurrences of a given motif (above the threshold score) within 500 bp upstream of the predicted ATG start codon. Two other negative groups (carbohydrate synthesis genes \[[@pbio-0020352-Kim2]\] and a set of 194 randomly selected genes) yielded similar results (data not shown). "Enhancer Activity" is the relative strength of expression generated by a motif present in three copies upstream of the Δpes-10::GFP::HIS2B reporter. ND, not determined. ::: ![](pbio.0020352.g003) ::: Five of the remaining seven motifs were associated with either the Ph-E or Ph-L temporal expression groups ([Figure 3](#pbio-0020352-g003){ref-type="fig"}). The Early-1 and Early-2 motifs were found by screening promoters of the Ph-E gene group and occurred more frequently in Ph-E promoters (44% and 54% of promoters, respectively) than in Ph-L promoters (26% and 21% of promoters, respectively). Conversely, the Late-1, Late-2, and Late-3 elements were found in Ph-L promoter searches and were more likely to occur in Ph-L promoters (53%, 35%, and 15% of promoters) than in Ph-E promoters (27%, 16%, and 8% of promoters). Two other motifs named P-1 and P-2 (for pharyngeal gene motifs 1 and 2) were discovered by screening Ph-E and Ph-L promoters together. These elements occurred in Ph-E and Ph-L genes with comparable frequency and were more frequently represented in promoters of pharyngeal genes (P-1 in 34% and P-2 in 68% of promoters) than of non-pharyngeal control genes (P-1 in 21% and P-2 in 48% of promoters). These data suggest that P-1 and P-2 are pharyngeal regulatory elements that are not associated with temporal control. We reexamined Ph-E genes for additional elements present in genes that contained neither Early-1 nor Early-2 elements. Given that 48% of Ph-E genes contained one or more Early-1 or Early-2 elements ([Table S4](#st004){ref-type="supplementary-material"}), we examined the remaining 52% of Ph-E genes for the presence of additional motifs. This survey identified a single motif that appeared to be a variant of the Early-1 motif and was therefore named "E1var." As with Early-1, E1var was enriched in the promoters of Ph-E pharyngeal genes compared to Ph-L pharyngeal and non-pharyngeal genes ([Figure 3](#pbio-0020352-g003){ref-type="fig"}). No other motifs were identified above background in these follow-up searches. We performed a similar analysis of Ph-L genes that did possessed neither Late-1 nor Late-2 elements, but found no additional motifs. These results suggest that if other temporal elements exist, they are represented by relatively degenerate sequences, are shared by small numbers of genes, occur in both pharyngeal and non-pharyngeal genes, or are outside the 500-bp regions examined. In Vivo Activity of Candidate Pharyngeal Regulatory Elements {#s2d} ------------------------------------------------------------ Two tests demonstrated that six of the candidate motifs had biological activity. First, we performed "enhancer assays" to determine whether a motif was sufficient to activate expression when introduced into a heterologous basal promoter. Second, we used site-directed mutagenesis to inactivate a motif within a native pharyngeal promoter and examine whether it was necessary for expression. We used the Δpes-10 promoter for the heterologous enhancer assays ([Figure 4](#pbio-0020352-g004){ref-type="fig"}). This promoter does not activate GFP ([Figure 4](#pbio-0020352-g004){ref-type="fig"}D--[4](#pbio-0020352-g004){ref-type="fig"}F) but is competent to respond to enhancers in most or all tissues ([@pbio-0020352-Seydoux1]; [@pbio-0020352-Fire2]). Previous studies established that the PHA-4 binding site and the CEH-22 binding site could activate expression of Δpes-10::GFP in pharyngeal and pharyngeal muscle cells, respectively ([@pbio-0020352-Kuchenthal1]; [@pbio-0020352-Vilimas1]). We therefore used this reporter to determine whether our motifs could function as pharyngeal enhancers. ::: {#pbio-0020352-g004 .fig} Figure 4 ::: {.caption} ###### Five Newly Identified Motifs Function as Pharyngeal Enhancers (A--C) Nomarski differential contrast interference images of embryos representing three different stages of embryonic development: (A) "early" development, when the pharynx primordium is formed, (B) "mid" development, when the pharynx has completed cell division and attached to the presumptive buccal cavity, and (C) "late" development, when pharynx development is almost complete and the embryo is about to hatch. Images on the left are of "early" embryos, images in the middle are of "mid" embryos, and images on the right are of "late" embryos. (D--U) Representative transgenic embryos showing expression from reporter constructs containing the Δpes-10 promoter alone (D--F) or with insertion of three copies of Early-1 (G--I), Early-2 (J--L), Late-2 (M--O), P-1 (P--R), or P-2 (S--U). Dashed lines indicate the outline of the developing pharynx. ::: ![](pbio.0020352.g004) ::: Our heterologous promoter assay demonstrated that five of the seven pharyngeal motifs could function as enhancers in vivo. Early-1 and Early-2 activated pharyngeal expression early, after specification of pharyngeal precursors at the 200-cell stage but before formation of the pharynx primordium ([Figure 4](#pbio-0020352-g004){ref-type="fig"}G--[4](#pbio-0020352-g004){ref-type="fig"}L). In both cases, activity was confined to embryos. Early-1 was active in most or all pharyngeal cells, with occasional activity in non-pharyngeal cells. Early-2 was active in most or all pharyngeal cells and many non-pharyngeal cells in the head, suggesting that Early-2 could be a regulator of anterior or head-specific expression. The Late-2 element enhanced expression in the majority of pharyngeal cells, beginning at mid-embryogenesis, when the pharynx primordium initiates morphogenesis and differentiation ([Figure 4](#pbio-0020352-g004){ref-type="fig"}M--[4](#pbio-0020352-g004){ref-type="fig"}O; 1.5 fold stage; [@pbio-0020352-Sulston2]; [@pbio-0020352-Portereiko1]). Late-2-dependent activity continued through embryogenesis but was relatively low or absent in larvae and adults. P-1 and P-2 functioned in pharyngeal cells after formation of the pharynx primordium, but both were relatively weak, variable enhancers compared to Early-1, Early-2, and Late-2 ([Figure 4](#pbio-0020352-g004){ref-type="fig"}P--[4](#pbio-0020352-g004){ref-type="fig"}U). Late-3 exhibited no enhancer activity. E1var, Early-3, and P-3 were not tested with this assay because of their similarity to other motifs (Early-1, CEH-22, and PHA-4, respectively). We conclude that five of the candidate elements can activate transcription broadly within the pharynx at distinct developmental stages. Tests of the Late-1 element revealed that this motif functions as a repressor rather than an enhancer ([Figure 5](#pbio-0020352-g005){ref-type="fig"}). We demonstrated repression using a modified version of the Δpes-10 promoter in which three copies of the PHA-4 binding site were placed upstream of the Δpes-10 basal promoter to activate expression within the digestive tract (3×pha-4P::Δpes-10::GFP). This construct expressed robustly in embryos and weakly in larvae and adults. To examine the Late-1 element for repressive activity, we inserted three copies of the Late-1 motif upstream of the PHA-4 sites. Remarkably, the presence of the Late-1 elements resulted in a marked decrease in embryonic GFP expression but had no effect on larval expression ([Figure 5](#pbio-0020352-g005){ref-type="fig"}). Embryos that expressed GFP from the 3×Late-1::3×pha-4P::GFP reporter were late-stage embryos in which the pharynx was nearly or completely developed. Conversely, the Late-1 element was not active when tested for enhancer activity with the Δpes-10 promoter (data not shown). Thus, Late-1 functioned as a repressor of early gut expression. Late-1 may also function in cells outside of the digestive tract, but this aspect of regulation was not tested. ::: {#pbio-0020352-g005 .fig} Figure 5 ::: {.caption} ###### Late-1 Represses Early PHA-4-Dependent Expression Percent values indicate the percentage of transgenic animals exhibiting pharyngeal GFP expression. A reporter construct with three copies of a high-affinity PHA-4 site (TGTTTGC) upstream of the Δpes-10 promoter (A) expresses GFP in pharyngeal cells of most transgenic embryos (B) and roughly one-third of transgenic larvae (C). The addition of three copies of the Late-1 element from R07B1.9 (CCTTGGCGGCGC) to this transgene (D) drastically reduces expression in transgenic embryos (E) but has no observable effect on transgenic larvae (F). Dashed lines indicate the outline of the pharynx. ::: ![](pbio.0020352.g005) ::: To characterize the temporal elements further, we examined their activity in endogenous pharyngeal promoters. We searched for examples of these motifs in predicted promoters that were conserved between C. elegans and C. briggsae to enrich for functionally relevant versions of motifs, rather than those that arose by chance alone. For Early-1 and Early-2, we tested the activity of these elements in the promoter of the early-onset pharyngeal gene K07C11.4. For Late-1, we tested the activity of this element in the promoter of the late-onset pharyngeal gene R07B1.9. We constructed a GFP reporter for K07C11.4 that faithfully recapitulated the expression observed by in situ hybridization experiments ([Figure 6](#pbio-0020352-g006){ref-type="fig"}; [@pbio-0020352-Kohara1]). K07C11.4::GFP was expressed from the stage of pharynx primordium formation to adulthood in the pharynx, midgut, and hindgut. K07C11.4::GFP was also active in the proximal somatic gonad of late larvae and adults. Alignment of K07C11.4 promoter sequences (500 bp upstream of the predicted ATG) between C. elegans and C. briggsae revealed stretches of conserved sequences. These regions contain two predicted PHA-4 binding sites (at −151 to −157 and −141 to −147, relative to the ATG), an Early-1 motif (−114 to −123), and an Early-2 motif (−217 to −225). The distal predicted PHA-4 site had a relatively high affinity for PHA-4 in vitro, consistent with this gene being expressed early in pharyngeal development ([@pbio-0020352-Gaudet1]). ::: {#pbio-0020352-g006 .fig} Figure 6 ::: {.caption} ###### Early-1 and Early-2 Elements Are Required for K07C11.4 Expression \(A) A portion of the promoter sequence of K07C11.4 from C. elegans (bottom) aligned with its ortholog from C. briggsae (top). Boxed regions show conserved predicted PHA-4 binding sites and Early-1 and Early-2 elements. Site-directed mutations that disrupt Early-1 and Early-2 ("E2 + E1 Mut") are shown below their respective wild-type ("E2 + E1 WT") sequence from K07C11.4. (B--E) Confocal images of mid-stage embryos expressing GFP under the control of the wild-type K07C11.4 promoter (B) or promoters with a mutation in Early-1 (C), Early-2 (D), or both Early-1 and Early-2 (E). Percentages are the fraction of transgenic embryos expressing GFP; the remainder of embryos do not express GFP. \(F) Expression of the wild-type K07C11.4 reporter in a subset of somatic gonad cells in an L4 animal (arrowheads). \(G) Mutation of the Early-1 element eliminates gonadal expression but does not strongly affect expression in other tissues, such as intestinal cells (arrows). Dashed lines indicate the outline of the developing pharynx. ::: ![](pbio.0020352.g006) ::: Early-1 and Early-2 elements were both positive regulatory elements for K07C11.4. Mutation of either Early-1 or Early-2 elements resulted in a decrease in the strength of reporter expression, but did not affect timing or cell-type specificity ([Figure 6](#pbio-0020352-g006){ref-type="fig"}C, [6](#pbio-0020352-g006){ref-type="fig"}D, and [6](#pbio-0020352-g006){ref-type="fig"}G). Interestingly, the two elements made distinct contributions to the activity of the K07C11.4 promoter: mutation of the Early-2 element had a relatively moderate effect on all aspects of K07C11.4 promoter activity, while mutation of the Early-1 element had only a mild effect on expression in the digestive tract but completely abolished expression in the somatic gonad. Simultaneous mutation of both elements virtually extinguished expression in all tissues, suggesting that the two elements functioned additively for K07C11.4 expression ([Figure 6](#pbio-0020352-g006){ref-type="fig"}E). We conclude that Early-1 and Early-2 elements are bona fide cis-regulatory elements for early gene expression in the pharynx and some non-pharyngeal tissues. To assess Late-1 activity, we constructed a reporter for R07B1.9 that reproduced the endogenous pattern of expression (from in situ hybridizations; [@pbio-0020352-Kohara1]; [Figure 7](#pbio-0020352-g007){ref-type="fig"}). R07B1.9 was activated during the terminal stages of pharynx development, (the 3-fold stage; [@pbio-0020352-Sulston2]), and its expression was maintained throughout the life of the animal ([Figure 7](#pbio-0020352-g007){ref-type="fig"}B and [7](#pbio-0020352-g007){ref-type="fig"}C). The promoter of R07B1.9 contained two predicted PHA-4 binding sites (−243 to −249 and −221 to −227) and one Late-1 element (−170 to −180), all of which were conserved in C. briggsae ([Figure 7](#pbio-0020352-g007){ref-type="fig"}A). ::: {#pbio-0020352-g007 .fig} Figure 7 ::: {.caption} ###### The Late-1 Element Negatively Regulates R07B1.9 \(A) A portion of the promoter sequence of R07B1.9 from C. elegans (bottom) aligned with its ortholog from C. briggsae (top). Boxed regions show conserved predicted PHA-4 binding sites and a Late-1 element. The site-directed mutation that disrupts Late-1 ("Mut") is shown below the respective wild-type sequence from R07B1.9. (B and C) Confocal images of representative early and late embryos expressing GFP under the control of the wild-type R07B1.9 promoter. (D and E) Confocal images of representative early and late embryos expressing GFP under the control of the R07B1.9 promoter with a mutation in the Late-1 element. Note the early activation of R07B1.9 when Late-1 is inactivated. Dashed lines indicate the outline of the developing pharynx. ::: ![](pbio.0020352.g007) ::: The Late-1 element functioned as a repressor for R07B1.9, as predicted. Elimination of the Late-1 element led to precocious expression 2--3 h earlier than the wild-type ([Figure 7](#pbio-0020352-g007){ref-type="fig"}D and [7](#pbio-0020352-g007){ref-type="fig"}E). GFP was first visible when the pharynx primordium was formed and continued throughout the life of the animal. The mutant reporter was expressed in the same cells and at the same approximate strength as the wild-type, suggesting that disruption of the Late-1 element specifically affected the onset of gene expression. We conclude that six of the motifs (Early-1, Early-2, Late-1, Late-2, P-1, and P-2) display activity in vivo and that at least three of these contribute to temporal regulation of endogenous pharyngeal genes. Temporal Elements and PHA-4 Sites Determine Onset of Expression {#s2e} --------------------------------------------------------------- Our analyses suggested that the Early and Late elements function in combination with PHA-4 sites to modulate the onset of pharyngeal gene expression. We performed three tests to address the generality of this model. First, we surveyed the Ph-E and Ph-L gene clusters to determine the configurations of the five regulatory elements (Early-1, Early-2, Late-1, Late-2, and PHA-4 sites) within endogenous promoters. Second, we constructed synthetic promoters to examine the interplay between the temporal elements and PHA-4 binding sites. Third, we combined our observations from the synthetic and endogenous promoters to search the genome for genes containing similar configurations of the five cis-regulatory motifs. ### Temporal regulatory elements in Ph-E and Ph-L genes {#s2e1} We examined Ph-E and Ph-L genes to see if their promoters contained distinct combinations of Early, Late, and PHA-4 sites ([Table 4](#pbio-0020352-t004){ref-type="table"}; [Table S4](#st004){ref-type="supplementary-material"}). For PHA-4 sites, we also examined the predicted affinity of these sites because the affinity of PHA-4 for its binding sites influences the onset of expression ([@pbio-0020352-Gaudet1]). For Ph-E and Ph-L, 22/33 and 19/33 genes, respectively, had temporal elements and predicted PHA-4 sites that were conserved in their C. briggsae orthologs. Interestingly, particular combinations of elements appear to be associated with Ph-E or Ph-L genes ([Table 4](#pbio-0020352-t004){ref-type="table"}). For example, combinations of Early elements with PHA-4 sites of predicted high or medium affinity are far more frequent in Ph-E genes than in Ph-L genes (11/22 \[50%\] versus 2/19 \[11%\], respectively), suggesting that this configuration of elements promotes early pharyngeal expression. In contrast, combinations of Late elements with PHA-4 sites of varying affinity are frequent in Ph-L genes (9/19 \[47%\]) but do not occur at all in Ph-E genes. In addition, promoters with only low-affinity PHA-4 sites together with any temporal elements are more common in Ph-L than in Ph-E genes (4/19 \[21%\] versus 2/22 \[9%\], respectively). These trends suggest that late pharyngeal gene expression is promoted by the combination of either Late elements with any PHA-4 site, or low-affinity PHA-4 sites with any temporal elements. We conclude that expression of at least half of Ph-E and Ph-L genes can be accounted for by a combination of PHA-4 sites and temporal elements. ::: {#pbio-0020352-t004 .table-wrap} Table 4 ::: {.caption} ###### Combinations of Temporal Elements and PHA-4 Sites Predict Onset of Pharyngeal Gene Expression ::: ![](pbio.0020352.t004) For temporal elements, "Early" indicates the presence of one or more of Early-1 and/or Early-2, and "Late" indicates the presence of one or more of Late-1 and/or Late-2. In all cases, motifs are present in both the C. elegans and C. briggsae orthologs of a gene. PHA-4 sites are as follows: "High" is TGTTTGC, TGTTGAC, or TGTTTAC; "Med" is TGTTTGT; "Low" is TATTTGT ([@pbio-0020352-Gower1]; [@pbio-0020352-Gaudet1]) (N. Gower and H. Baylis, personal communication); "Unk" is all other TRTTKRY except TATTTAT, which shows no appreciable binding of PHA-4 in vitro (unpublished data). We predict onset of expression for particular combinations of elements based on the prevalence in the promoters of Ph-E and Ph-L genes ^a^ Onset of expression of these genes will depend on the relative affinity of the PHA-4 sites. For a complete description of individual genes and their elements, see Table S4 ::: ### Synthetic promoters recapitulate temporal expression {#s2e2} The combination of elements observed in natural promoters suggests that a gene with a high-affinity PHA-4 site and an Early element will be expressed early in development, while a gene with a high-affinity PHA-4 site and a Late element will be expressed later in development. From the existing Ph-E and Ph-L promoters, however, the output of certain promoter configurations cannot be reliably predicted. For example, when a low-affinity PHA-4 site is paired with an Early element, is the gene activated early or late in development? A low-affinity PHA-4 site in the presence of an Early element might be expected to activate expression relatively late, as determined by the binding of PHA-4 to the low-affinity site, or early, if the Early element potentiates earlier expression. To test these ideas, we constructed artificial promoters within the context of the Δpes-10 promoter and examined their expression patterns. We first investigated the behavior of the high- and low-affinity PHA-4 binding sites in synthetic promoters, because the configuration of Ph-E and Ph-L promoters together with previous work ([@pbio-0020352-Gaudet1]) suggested that onset of expression was influenced by the affinity of PHA-4 sites. To test this idea, we compared the activity of three copies of either a high- or low-affinity PHA-4 site placed in front of the Δpes-10 promoter ([Figure 8](#pbio-0020352-g008){ref-type="fig"}). As observed previously ([@pbio-0020352-Okkema1]), three copies of a high-affinity PHA-4 site were sufficient to activate pharyngeal expression beginning early in development, prior to formation of the pharynx primordium (n = 34/36). In contrast, three copies of a low-affinity PHA-4 site were sufficient to activate pharyngeal expression later in development. Early expression was observed significantly less frequently with the low-affinity constructs than with the high-affinity constructs (n = 28/71). Notably, the strength of expression of both constructs was comparable in late embryos. We conclude that promoter activation depends on the affinity of PHA-4 for its binding sites and cannot be compensated for by multimers. This finding agrees with data that show FoxA factors bind DNA as monomers, not cooperatively ([@pbio-0020352-Clark1]). ::: {#pbio-0020352-g008 .fig} Figure 8 ::: {.caption} ###### High-Affinity PHA-4 Sites Activate Pharyngeal Expression Earlier Than Low-Affinity Sites Percent values indicate the percentage of transgenics exhibiting pharyngeal GFP expression. Dashed lines indicate the outline of the developing pharynx. (A--C) A reporter construct with three copies of a high-affinity PHA-4 site (TGTTTGC) upstream of the Δpes-10 promoter reproducibly activates pharyngeal expression from the time of pharynx primordium formation ("early") through embryogenesis. (D--F) A reporter construct with three copies of a low-affinity PHA-4 site (TATTTGT) upstream of the Δpes-10 promoter activates pharyngeal expression from the time of attachment of the pharynx to the mouth ("mid") through embryogenesis. ::: ![](pbio.0020352.g008) ::: We next tested synthetic promoters containing a single PHA-4 site (either high or low affinity) with a single temporal element (either Early-1 or Late-2) to determine how these elements behave in combination. The synthetic promoter constructs differed from our initial Δpes-10::GFP reporter survey in that they resembled the configuration of endogenous promoters. While temporal elements and PHA-4 sites do not appear to exhibit conserved spacing or order within pharyngeal promoters, each element typically occurs in one or two copies per promoter. We therefore constructed reporters in which one copy of either an Early-1, Early-2, or Late-2 element was paired with one copy of either a high- or low-affinity PHA-4 site, separated by approximately 100 bp. The onset of expression for artificial promoters containing an Early-1 site depended on the relative affinity of the accompanying PHA-4 site ([Figure 9](#pbio-0020352-g009){ref-type="fig"}), as predicted. The promoter with a high-affinity PHA-4 site was reliably expressed at all embryonic stages examined, including early pharyngeal development (n = estimated number of transgenic embryos scored = 17/23; see [Materials and Methods](#s4){ref-type="sec"} for an explanation of the estimate). In contrast, the promoter with the low-affinity PHA-4 site was consistently expressed in late embryos, but only infrequently in earlier stages (n = 4/58). Because both artificial promoter constructs activated expression comparably in late embryos, we conclude that the differences in early expression reflect a genuine difference in the onset of expression between constructs rather than a difference in the strength or penetrance of expression. These data support a model in which the Early-1 element is able to mediate early pharyngeal gene expression but where the onset of gene expression is ultimately limited by the relative affinity of the PHA-4 binding site. We similarly tested Early-2 together with a high-affinity PHA-4 site, but did not observe any expression above background, suggesting that this combination or configuration of sites was not sufficient to activate pharyngeal expression. ::: {#pbio-0020352-g009 .fig} Figure 9 ::: {.caption} ###### Temporal Elements Combined with PHA-4 Sites Regulate the Onset of Pharyngeal Expression "Early," "mid," and "late" are as defined in [Figure 3](#pbio-0020352-g003){ref-type="fig"}. E1, Early-1; L2, Late-2; High, high-affinity PHA-4 site (TGTTTGC); Low, low-affinity PHA-4 site (TATTTGT). Percent values indicate the percentage of transgenics exhibiting pharyngeal GFP expression; the remainder of embryos do not express GFP. A reporter construct with one copy of the Early-1 element and one copy of a high-affinity PHA-4 site is expressed in "early" to "late" embryos (A--C). In contrast, a reporter with one copy of the Early-1 element and one copy of a low-affinity PHA-4 site is not consistently expressed until the "mid" to "late" stages (D--F). Reporters with one copy of Late-2 and one copy of either a high-affinity (G--I) or low-affinity (J--L) PHA-4 site are expressed in "mid" and "late" stage embryos. Dashed lines indicate the outline of the developing pharynx. ::: ![](pbio.0020352.g009) ::: Artificial promoters containing the Late-2 element were not expressed until mid-embryonic to late embryonic development, regardless of the relative affinity of the PHA-4 site ([Figure 9](#pbio-0020352-g009){ref-type="fig"}; n = 1/73 early embryos for constructs with a high-affinity PHA-4 site, n = 2/37 early embryos for constructs with a low-affinity site). Because the Late-2 element behaved as an enhancer, we hypothesize that the factor or factors that act through Late-2 are not available until late in development and that their absence in early development results in delayed expression of Late-2 dependent genes. ### Genome searches identify additional pharyngeal genes {#s2e3} The temporal motifs offered an opportunity to discover new pharyngeal genes based solely on predicted cis-regulatory sites. We searched the C. elegans and C. briggsae genomes for conserved occurrences of the temporal motifs, together with PHA-4 sites, to determine whether combinations of these sites were predictive for pharyngeal genes. We required the presence of three elements within a 500-bp stretch because Ph-E and Ph-L averaged a total of 3.7 elements in their promoters and because artificial promoters with only two elements generated weaker expression than endogenous promoters ([Figure 9](#pbio-0020352-g009){ref-type="fig"}). We chose a stringent threshold score for each element that maximized the ability to distinguish elements in control runs of early versus late pharyngeal genes (see [Materials and Methods](#s4){ref-type="sec"}). Although these thresholds reduced the number of elements identified in the positive control set (e.g., the Early-1 element in Ph-E promoters) they also dramatically lowered the false positive rate (e.g., the Early-1 element in Ph-L promoters). The combination of Early-1, Early-2, and PHA-4 sites provided a powerful approach to predict early pharyngeal genes. Of 40 genes with conserved copies of all three elements, 20 had available expression data, and 70% of these were expressed in the pharynx (14 genes; [Table S5](#st005){ref-type="supplementary-material"}). In contrast, of 194 randomly selected genes, 57 had expression data and only 12% (7/57) of these were expressed in the pharynx (data not shown), indicating that the results of the genomic searches are significantly enriched for pharyngeal genes (p = 6 × 10^−7^). Furthermore, of the 14 pharyngeal genes identified by the genomic search, 86% (12/14) were expressed early in pharyngeal development, as predicted. The 14 pharyngeal genes included ten genes not identified in our microarray experiments, reflecting the power of bioinformatics in predicting gene expression. Given the possibility that E1var is a functional temporal element, we performed genomic searches for genes that contain conserved E1var plus Early-2 plus PHA-4 elements together within 500 bp of their predicted start codon. This search identified 120 genes with some overlap with the Early-1 search output. Of these 120 genes, 53 had expression data ([Table S5](#st005){ref-type="supplementary-material"}), and 47% of these (25/53) were expressed in the pharynx, a significant enrichment compared to the random set (p = 6 × 10^−5^), suggesting that, like Early-1, E1var in combination with other sites is predictive for pharyngeal expression. Of these pharyngeal genes, 70% (16/23; onset of expression not determined for two genes) were expressed early, as predicted. Combining the results of the Early-1 and E1var searches, we identified a total of 35 pharyngeal genes, 73% of which (24/33) were expressed early. For late expression, 61 genes had conserved Late-1, Late-2, and PHA-4 sites, and these were also enriched for pharyngeal genes. Thirty of these genes had expression data, of which 33% (ten genes) were pharyngeally expressed ([Table S5](#st005){ref-type="supplementary-material"}), showing significant enrichment compared to the random gene set (p = 0.02). Strikingly, the Late elements accurately predicted the timing of expression in the pharynx, as all ten pharyngeal genes were expressed late in pharyngeal development. Considering the Early and Late genomic searches together, we identified 45 pharyngeal genes, 36 of which were not present in our microarray positives. Furthermore, the onset of expression of these 45 pharyngeal genes was predicted with 88% overall accuracy. There are three different reasons for our microarray experiments not identifying the 36 new pharyngeal genes: (1) some genes are expressed in both pharyngeal and non-pharyngeal cells and would not be significantly enriched in the par-1 versus skn-1 samples, (2) some genes are expressed post-embryonically and would not have been present in the embryonic samples used for the microarray experiments, and (3) some genes scored just below the 2-fold enrichment threshold for inclusion in our positive set. Discussion {#s3} ========== The advent of microarray technology and bioinformatics has provided a powerful tool to dissect the transcriptional regulatory circuits that guide complex developmental processes. We have analyzed pharynx organogenesis and defined novel regulatory elements that contribute to temporal gene expression. Our screens differ from previous examples of bioinformatic analyses in that the regulatory motifs were identified in unbiased searches of promoter sequences and subsequently tested for biological activity. We used three assays to demonstrate a function in vivo. First, we tested whether the regulatory elements were necessary for expression of native pharyngeal genes. Second, we determined whether they were sufficient for pharyngeal expression from synthetic promoters. And third, we did genome-wide searches based on these elements, a process that identified 19 new pharyngeal genes, of which 88% displayed the expected onset of expression. The cis-regulatory motifs discovered here, combined with the PHA-4 binding site, establish a regulatory network that can account for the timing of activation of at least half of C. elegans pharyngeal genes. A Model for Temporal Control of Pharyngeal Gene Expression {#s3a} ---------------------------------------------------------- We previously proposed a model in which the relative affinity of PHA-4 for its binding sites controls the onset of pharyngeal gene expression. Because PHA-4 protein levels increase during development, we proposed that initially PHA-4 levels are low and only high-affinity sites are sufficiently occupied by PHA-4 to result in gene activation. As PHA-4 levels increase over time, lower-affinity sites also become occupied, leading to the expression of those genes. In addition, binding site affinity likely affects PHA-4 occupancy even at stable PHA-4 concentrations. The degree of PHA-4 occupancy would likely alter the probability of productive transcription of a target gene. However, the affinity model could not explain all temporal expression because some late-expressed pharyngeal genes had high-affinity PHA-4 binding sites (e.g., myo-2 and R07B1.9). In this study, we have identified additional regulatory elements that establish the onset of pharyngeal gene expression in combination with PHA-4. We suggest that the affinity of PHA-4 for its binding site determines the earliest possible time of pharyngeal gene expression but that other factors must be present for a promoter to be active. The Early elements likely represent binding sites for transcription factors that are available throughout embryonic pharyngeal development. In contrast, the Late-2 element is probably recognized by a transcription factor that is not available until midway through pharynx development, thereby delaying expression of these pharyngeal genes regardless of the quality of the PHA-4 binding sites. We note that our searches were restricted to genes with similar onset of expression, in order to identify temporal regulatory elements, and there may be yet more undiscovered elements that control temporal expression. For example, an element that confers cell-type-specific expression could also control onset of expression according to the availability of the relevant binding factor. Consistent with the idea of additional motifs, the synthetic promoters with two enhancer elements were functional but weak activators of expression compared to native promoters, which likely contain more regulatory motifs (e.g., myo-2). In addition to enhancers of pharyngeal expression, we identified one negative regulatory element, Late-1, that is dominant to PHA-4 in early pharynx development. The proximal PHA-4 consensus site in the R07B1.9 promoter is predicted to be a high-affinity binding site, suggesting that R07B1.9 should be expressed early in pharynx development, not late as is observed. Mutation of the Late-1 element enabled the R07B1.9 promoter to fire at earlier developmental stages. We hypothesize that some factor binds to Late-1 and represses expression in early pharynx development. Subsequently, the Late-1 factor is presumably inactivated or downregulated to permit gene expression. The elements characterized here can account for the temporal expression patterns of many but not all known pharyngeal genes. One possible reason for this is that we do not yet have sufficient information to identify all functionally relevant occurrences of these elements. For example, some occurrences of the temporal elements may not be biologically meaningful, thereby generating false positives, while others may be hard to identify because of sequence heterogeneity or placement within cis-regulatory sequences. Our analyses initially focused on regions within 500 bp of the predicted start codon of genes, but we further searched −501 to −1,000 upstream of the ATG for Ph-E and Ph-L genes whose expression could not be accounted for by elements in the region of −1 to −500. However, these extended searches did not find any conserved elements that could further account for onset of expression (data not shown). Another possibility is that regulatory elements may be found within introns or the 3′ UTR ([@pbio-0020352-Shibata1]; [@pbio-0020352-Marshall1]; [@pbio-0020352-Gaudet1]; [@pbio-0020352-Kirouac1]). Consistent with this observation, we find that 40% of the Ph-E and Ph-L genes show significant conservation of intron sequence between C. elegans and C. briggsae (data not shown). Thus, some of the genes whose onset of expression cannot be accounted for in our analyses likely have larger and more complex regulatory regions than the 500-bp window we used. Nonetheless, this limited sequence window allowed us to account for the onset of expression of roughly half of the Ph-E and Ph-L genes. An interesting feature of the Early-1 element is that it is necessary for the expression of K07C11.4 in the proximal somatic gonad, as well as in the pharynx. Some, though not all, genes containing conserved Early-1 elements are expressed in the proximal somatic gonad (e.g., F30H5.3, F47D12.7, and ncr-1; [@pbio-0020352-Kohara1]), consistent with Early-1 being necessary but not sufficient for gonadal expression. PHA-4 is also expressed in the somatic gonad, including the proximal region ([@pbio-0020352-Azzaria1]; [@pbio-0020352-Kalb1]). The finding of multiple genes expressed in the two tissues under control of Early-1 and possibly PHA-4 suggests the presence of shared regulatory mechanisms in the two organs. Both the pharynx and proximal somatic gonad are epithelial tubes that connect to the external environment and as such may have conserved features that are regulated by some of the same factors. The role of PHA-4 in the gonad has not been carefully analyzed, but PHA-4 is critical for proper gonad development post-embryonically ([@pbio-0020352-Ao1]). Candidate trans-Acting Factors {#s3b} -------------------------------- The identification of new pharyngeal regulatory elements provides an entry point for identifying the relevant transcription factors that bind these sequences. We searched the TRANSFAC database and available literature for possible matches to the temporal elements identified here ([@pbio-0020352-Knuppel1]; [@pbio-0020352-Matys1]). While we did not discover obvious candidate factors for the relevant transcription factors, we did find intriguing similarities between our elements and known transcription factor binding sites. The Early-1 element resembles the recognition sequence for Drosophila ZESTE and GAGA (YGAGYG and GAGAG, respectively; [@pbio-0020352-Benson1]; [@pbio-0020352-Omichinski1]). However, no clear ortholog of either gene exists in C. elegans. Early-2 resembles a hemi-site for nuclear hormone receptors, but the identification of specific candidate factors is complicated by the existence of 284 predicted nuclear-hormone-receptor-encoding genes in C. elegans ([@pbio-0020352-Maglich1]). The Late-1 element is a GC-rich sequence that resembles a Sp-factor binding site and behaves as a negative regulator of early expression. C. elegans has four Sp-like homologs ([@pbio-0020352-Zhao1]), one of which is a predicted pharyngeal gene (F45H11.1). However, F45H11.1(RNAi) did not affect expression of the Late-1-regulated gene R07B1.9 (data not shown). The Late-2 element (TTTTTCC) most closely resembles a Dorsal/Rel-homology domain binding site (KGGWWWWCCC; [@pbio-0020352-Matys1]), but there are no C. elegans Rel-homology domain proteins. Given the lack of other obvious candidate factors for the temporal elements, molecular or genetic screens will be required to identify the relevant transcription factors. Elements and Regulatory Modules {#s3c} ------------------------------- Recent studies of transcription factor binding sites have revealed cases of multimers of a single site being important for expression ([@pbio-0020352-Berman1]; [@pbio-0020352-Markstein1]; [@pbio-0020352-Yoo1]) and cases of entire modules of elements being conserved features of some promoters ([@pbio-0020352-Senger1]). By contrast, we see little evidence of conserved spacing, order, or organization of our elements within pharyngeal promoters. Genomic searches for genes with multiple copies of a single element yielded few pharyngeal genes, suggesting that the temporal elements and PHA-4 sites typically act in single copy (data not shown). The analysis of PHA-4 targets suggests two strategies of transcriptional control. We propose that a minority of target genes respond consistently to the presence of PHA-4. For example, K07C11.4, T05E11.3, and M05B5.2 are active broadly throughout the pharynx and in other cells that express PHA-4, such as the gonad or rectum ([@pbio-0020352-Gaudet1]). These promoters contain four or more predicted PHA-4 binding sites, including at least one high-affinity site. The density of high-quality PHA-4 binding sites may promote activation of these genes whenever PHA-4 is present. This strategy may have been adopted by other transcription factors. For example, likely target genes of the Notch effector CSL (also known as CBF1/RBP-Jκ~e~, Su(H), and LAG-1) have been discovered in C. elegans and Drosophila ([@pbio-0020352-Christensen1]; [@pbio-0020352-Rebeiz1]; [@pbio-0020352-Yoo1]). Many of these genes encode components of the Notch signaling pathway that function upstream of CSL and would therefore be expected to respond broadly to CSL as part of a regulatory feedback loop. Intriguingly, these genes contain multiple copies (e.g., 15--25) of the CSL binding site consensus, which may facilitate CSL-mediated activation ([@pbio-0020352-Christensen1]). On the other hand, the majority of pharyngeal genes respond to PHA-4 in some cellular environments but not others ([@pbio-0020352-Okkema1]; [@pbio-0020352-Gaudet1]). These genes appear to depend on a second regulatory strategy, in which combinations of elements synergize, with no individual element sufficient for transcription. These promoters typically carry one or two copies of any given element, as observed for the myo-2 promoter ([@pbio-0020352-Okkema1]; [@pbio-0020352-Thatcher1]; [@pbio-0020352-Gaudet1]; [@pbio-0020352-Ao1]). Similarly, CSL factors have target genes that are activated in only a subset of tissues, and these contain only one or a few high-quality binding sites (for example, vestigial or D-pax2; [@pbio-0020352-Kim1]; [@pbio-0020352-Flores1]). The combinatorial mode of regulation relies on the relatively poor transcriptional activity of the individual factors. For example, a single Early-1 or PHA-4 binding site cannot activate expression of Δpes-10::GFP. Moreover, PHA-4 can transform some cells towards a pharynx fate, but is not as potent as other developmental regulators such as end-1 ([@pbio-0020352-Horner1]; [@pbio-0020352-Kalb1]; [@pbio-0020352-Zhu2]). Ectopically-expressed END-1 can transform the entire embryo into midgut ([@pbio-0020352-Zhu2]). Accordingly, END-1 is expressed briefly within midgut precursors, where it likely activates a uniform panel of downstream targets ([@pbio-0020352-Zhu1]). We suggest that the inherent activity of a transcriptional regulator coupled with promoter architecture defines the range of target genes available to a developmental transcription factor. The combinatorial mode of regulation exhibited by PHA-4 provides the organism with transcriptional flexibility in two ways. First, it provides a mechanism for selectivity for Fox transcription factors. The worm genome encodes fifteen Fox transcription factors, and these proteins regulate diverse biological activities such as cell fate specification, longevity, and cell migration ([@pbio-0020352-Miller1]; [@pbio-0020352-Ogg1]; [@pbio-0020352-Horner1]; [@pbio-0020352-Kalb1]; [@pbio-0020352-Nash1]; [@pbio-0020352-Sarafi-Reinach1]; [@pbio-0020352-Hope2]). Combinatorial regulation affords the animal with a means to distinguish different target genes for different Fox proteins, all of which share a similar DNA binding domain. Second, the combinatorial strategy enables PHA-4 to play a broad role in the pharynx. PHA-4 is activated at the earliest stages of organogenesis in all pharyngeal cells, where it is required to specify different pharyngeal cell types. PHA-4 continues to be expressed throughout the life of the animal, where it likely controls pharyngeal function and growth. These different functions presumably reflect different target genes activated in different cell types or during different developmental stages. FoxA factors in other animals exhibit long-term expression and, like PHA-4, are poor inducers of cell fate when expressed ectopically ([@pbio-0020352-Sasaki1]). FoxA target genes such as albumin require additional factors for activation, and FoxA promoter association is not sufficient for transcription ([@pbio-0020352-Zaret1]). These data indicate that mammalian FoxA proteins likely rely on a transcriptional strategy similar to that of worms. Materials and Methods {#s4} ===================== {#s4a} ### Identification of genes selectively expressed in the pharynx {#s4a1} C. elegans strains KK822 (par-1(zu310) IV; [@pbio-0020352-Kemphues1]) and EU1 (skn-1(zu67)/DnT1 IV;V; [@pbio-0020352-Bowerman1]) were grown in liquid culture with OP50 as a food source, synchronized, and harvested. For KK822, we shifted synchronized homozygotes to the restrictive temperature (25 °C) and isolated embryos by hypochlorite treatment ([@pbio-0020352-Sulston1]). For EU1, we grew synchronized animals and plated young adults. skn-1/DnT1 worms are uncoordinated, while skn-1 homozygotes are non-uncoordinated, allowing us to enrich for skn-1 homozygotes using a plate crawling assay. To synchronize, we performed a first round of hypochlorite embryo isolation and then allowed embryos to hatch overnight in liquid culture lacking food to obtain L1 larvae. We then added food to the cultures and grew the animals for 2--3 d to age them. For par-1, animals were aged to young adults, and embryos were collected from these by hypochlorite treatment. For the skn-1/DnT1 strain, animals were aged to L4 and then transferred to large plates for the crawling assay and later collected as young adults for embryo isolation. Collected embryos from either strain were aged another 3--6 h in small liquid cultures to ensure that par-1 and skn-1 embryos were at approximately the same stage of development, as determined by examining a sample of embryos under the light microscope. To extract total RNA from isolated C. elegans embryos, embryos pellets were frozen in microfuge tubes, crushed with a plastic pestle, and resuspended in RNA extraction buffer (1% lauroyl sarcosine, 0.1 M Tris base, 0.1 M NaCl, and 20 mM EDTA), followed by several rounds of phenol:chloroform extraction and ethanol precipitation ([@pbio-0020352-Horner1]; [@pbio-0020352-Gaudet1]). Once-selected poly-A+ RNA was purified using the PolyTract Isolation Kit (Promega, Madison, Wisconsin, United States). par-1 cDNAs were labeled with Cy3, and skn-1 cDNAs were labeled with Cy5. Labeled cDNA was prepared from 5 μg of poly-A+ RNA by the Huntsman Cancer Institute Microarray Core Facility. Construction and probing of the microarrays was as described by Reinke et al. under the auspices of the Kim lab ([@pbio-0020352-Reinke1]). ### Analysis of microarray data {#s4a2} We performed two microarray experiments using microarrays containing 62% of the C. elegans genome ("partial arrays," experiments PS1 and PS2) and three experiments using microarrays containing 94% of the genome ("full arrays," experiments PS3, PS4, and PS5). Our previous microarray experiments (PS1--PS3) detected 242 positives, corresponding to 227 genes ([@pbio-0020352-Gaudet1]). To extend the identification of candidate pharyngeal genes, we included data from PS4 and PS5. In this case, we selected genes that had an average log~2~(par-1/skn-1) ≥ 1.00 in PS1--PS5 and were expressed above background in three of five experiments. In this selection, we observed considerable background of non-pharyngeal genes, primarily genes with maternally contributed transcripts or expression in pre-gastrulation embryos. We hypothesize that this background is the result of our par-1 embryos being harvested at an earlier stage than our skn-1 embryos in experiments PS4 and PS5. Because PS1--PS3 do not appear to exhibit this difference in staging, we applied an additional selection criterion to our data, requiring that all positives have a log~2~(par-1/skn-1) ≥ 0.58 on PS1, PS2, or PS3. This threshold reduces the inclusion of non-pharyngeal genes identified by PS4 and PS5 and identifies a total of 241 microarray positives. We further subtracted probable maternal genes from this set of positives using the C. elegans expression map of [@pbio-0020352-Kim2], leaving us with 118 microarray positives corresponding to 112 new candidate pharyngeal genes. As a validation of these experiments, we note that among these 112 new genes are 44 genes with known expression patterns, and 33/44 (75%) of these are pharyngeally expressed. This fraction of pharyngeal genes is comparable to the fraction of pharyngeal genes present in our set of positives from PS1--PS3 (81/97 \[84%\]). ### Examining sequence conservation {#s4a3} C. elegans and C. briggsae upstream and downstream sequences were extracted from the genome using the Ensembl EnsMart tool ([@pbio-0020352-Kasprzyk1]). For four genes, ENSEMBL did not provide orthologous C. briggsae sequences. In these cases we used the Intronerator Tracks Display ([@pbio-0020352-Kent1]) to obtain the C. briggsae sequence. Sequences were then aligned and visualized by VISTA ([@pbio-0020352-Mayor1]). We scored as conserved those regions of DNA that had 75% or greater identity over 50 or more base pairs. Alignments in [Figures 6](#pbio-0020352-g006){ref-type="fig"} and [7](#pbio-0020352-g007){ref-type="fig"} were obtained from the Tracks Display feature of Intronerator ([@pbio-0020352-Kent1]). ### Motif searches using Improbizer {#s4a4} We used the Improbizer program ([@pbio-0020352-Ao1]; available at <http://www.soe.ucsc.edu/approximately/kent/improbizer/>), which employs a variation of the expectation maximization algorithm ([@pbio-0020352-Bailey1]), to search for motifs. Improbizer was able to find expected regulatory elements (i.e., motifs that resemble PHA-4 and CEH-22 binding sites) in our promoter sequences, while other algorithms did not. Improbizer can be configured to simultaneously search for motifs on both strands of DNA, search for more than a single occurrence of a motif on each DNA sequence, and use a separate set of sequences for a background model. The source code for the Improbizer is freely available, and is the definitive reference for the details of the algorithm. In using Improbizer to search the Ph-E and Ph-L gene sets for possible regulatory motifs, we initially searched for motifs occurring once per sequence, with an initial scan through the first five sequences entered. For each gene set we ran searches three or more times, varying the order of input for genes in each run. Motifs presented here were obtained with searches for a motif size of six. Searches for motifs of larger sizes (8--20 bases) recurrently found variations of the motifs presented here. Other parameters of Improbizer were used at their default settings. For background sequences, we used three different sets of sequences (foreground sequence, and upstream sequences from each of two gene groups defined by [@pbio-0020352-Kim2]: neuronal genes and carbohydrate metabolism genes) and obtained similar output matrices in all cases. All matrices presented here were obtained using the input sequence (foreground) as the background. This mixing was performed to prevent output being biased towards sequence motifs found in only the first few input genes. The motifs presented here were reproducibly obtained independent of the order of input sequences. As an initial screen for motifs that were over-represented in our test sets, we performed control runs in which the input gene sequence was randomized and searched. All motifs presented here obtained Improbizer scores greater than the scores of ten or more control runs. ### Identification of motifs using Cluster-Buster {#s4a5} Cluster-Buster finds the best possible occurrence of a motif (or motifs) in a given sequence and therefore requires the establishment of a threshold score to determine which occurrences are likely to be meaningful ([@pbio-0020352-Frith1]). We established threshold scores for our motifs that maximized the ratio of "hits" in a positive versus a negative group, with a "hit" defined as a gene that contained an occurrence of a motif above the threshold. For example, we chose a threshold score for Early-1 that gave the greatest ratio of Ph-E hits to Ph-L hits. These thresholds were then applied to searches of C. elegans Ph-E and Ph-L genes and their C. briggsae orthologs to determine which genes contained Early and Late elements in both C. elegans and C. briggsae. Position within the promoter was not required for a motif to be considered "conserved." For [Tables 3](#pbio-0020352-t003){ref-type="table"} and [S4](#st004){ref-type="supplementary-material"}, the following parameters were used: the cluster score threshold (C) and gap parameter (g) for all motifs were zero and 35, respectively. The motif threshold (m) used for Early-1, Early-2, E1var, and Late-1 was six, for Late-2 was seven, and for PHA-4 was five. The random gene set referred to in [Table 3](#pbio-0020352-t003){ref-type="table"} was generated using a random number generator to select 200 genes from a complete list of all predicted C. elegans genes. Duplicates or splice isoforms of a gene were collapsed to a single selection, resulting in a total of 194 genes. For the genome searches, we searched for genes that contained a set of elements (e.g., Early-1 plus Early-2 plus PHA-4) within the first 500 bp of upstream sequence in both C. elegans and C. briggsae. We applied more stringent thresholds to the elements for these searches, to minimize the identification of non-pharyngeal genes. We optimized the Early element thresholds by searching the genome to identify known Ph-E and Ph-L genes and selecting the search parameters that maximized the ratio of Ph-E/Ph-L genes identified. The Cluster-Buster ([@pbio-0020352-Frith1]) parameters for Early-1 plus Early-2 combined were C = 3.5, m = 6, and g = 35. The PHA-4 parameters were C = 1.9, m = 6, and g = 35. For E1var plus Early-2 combined the parameters were C = 1, m = 5.5, and g = 35. The PHA-4 parameters were C = 2, m = 6, and g = 35. Using these thresholds, our genomic searches identified four known Ph-E genes but no Ph-L genes. The same approach was used to optimize parameters for genome searches with the Late elements, maximizing the ratio of Ph-L/Ph-E genes identified. The parameters for Late-1 plus Late-2 combined were C = 2, m = 6, and g = 35. The PHA-4 parameters were C = 2.5, m = 6, and g = 35. ### Construction of plasmids {#s4a6} To construct transcriptional reporters, we amplified promoter sequences from genomic N2 DNA using gene-specific primers that contained restriction endonuclease sites to facilitate cloning. PCR products were cloned into the vector pAP.10, which carries a GFP::HIS2B translational fusion, resulting in a nuclear-localized GFP ([@pbio-0020352-Gaudet1]). This cloning strategy removed the pes-10 promoter sequence present in pAP.10. Enhancer constructs and synthetic promoters were built using synthetic oligonucleotides that were cloned into pAP.10, upstream of the Δpes-10 promoter fragment. Clones were verified by restriction digests and sequencing. For the triplicate enhancer sequences, we used the following insert sequences (sequences of the individual motifs are underlined; periods show spacing of elements in Late-1): Early-1, ATTATTATT; Early-2, T ACTTACTTACT; Late-1, ..C; Late-2, CTCCACTCCACTCCA; Late-3, ACTCACTCACTC; P-1, TTAATTAATTAA; P-2, TTTTTTTTT; PHA-4 (high), CTACCCCTACCCCTACCC; and PHA-4 (low), CTACCCCTACCCCTACCC. For the synthetic promoters, individual Early-1 or Late-2 sites were cloned in to the SphI and SalI sites of pAP.10, and individual PHA-4 sites were cloned in to the NheI and NsiI sites of pAP.10. Fragments from these single-site constructs were ligated to generate the constructs containing one temporal site and one PHA-4 site, with the sites separated by 95 bp of pAP.10 sequence. The sequences of the synthetic regions of these constructs were as follows (individual motifs are underlined): Early-1 plus PHA-4 (high), GCATGCTCGATTGTCGAC-(95-bp)-GCTAGCTACCCCCGGGATGCAT; Early-1 plus PHA-4 (low), GCATGCTCGATTGTCGAC-(95-bp)-GCTAGCTACCCCCGGGATGCAT; Late-2 plus PHA-4 (high), GCATGCTCGAGCTCCATCGAC-(95-bp)-GCTAGCTACCCCCGGGATGCAT; and Late-2 plus PHA-4 (low), GCATGCTCGAGCTCCATCGAC-(95-bp)-GCTAGCTACCCCCGGGATGCAT. Sequences chosen were the best match to the position weight matrix (PWM) generated by Improbizer, except Late-1 (which is based on the functional element in R07B1.9), P-1 (which is based on two overlapping PWMs), and the PHA-4 sites (which are based on a functional site in the Ce-pax-1 promoter; J. Stevenson and S. E. M., unpublished data). PWMs from Improbizer are listed in [Dataset S1](#sd001){ref-type="supplementary-material"}. Complete details of all oligonucleotides and plasmids are available upon request. ### Construction of transgenic lines {#s4a7} Several groups have reported artificial pharyngeal expression resulting from sequences present in the vector backbone of reporter constructs (e.g., [@pbio-0020352-Hope1]). To minimize this effect, we routinely removed all vector sequence from our reporter constructs prior to injection into C. elegans. For our transcriptional fusions, we used a gene-specific oligonucleotide together with an unc-54 3′ oligonucleotide that anneals downstream of the unc-54 3′ cassette present in our plasmids ([@pbio-0020352-Fire1]) to PCR-amplify linear fragments for injection. For our enhancer constructs, we used an oligonucleotide that anneals approximately 200 bp upstream of the MCS of pAP.10 ([@pbio-0020352-Gaudet1]) together with unc-54 3′ to amplify transgenes. We then digested the PCR products with either StuI or SphI to remove the remaining approximately 200 bp of vector sequence and gel-purified the desired fragment for microinjection. These adjustments ensured that there was no spurious expression in pharyngeal cells from Δpes-10::GFP in the absence of an enhancer or when three copies of a random sequence (corresponding to the degenerate sequence CWNCAYKGA) were placed in front of the Δpes-10 promoter ([Figure 4](#pbio-0020352-g004){ref-type="fig"}; data not shown). Linear transcriptional reporters were injected at 0.5--1.0 ng/μl together with 30 ng/μl pRF4 ([@pbio-0020352-Mello1]) cut with EcoRI and 70 ng/μl sheared herring sperm DNA ([@pbio-0020352-Kelly1]). In all cases where expression of transgenes was compared, injections were performed under the same conditions. To establish transgenic lines, we picked Roller animals from the F2 generation. For all transgenes, a minimum of two independent lines were analyzed. ### Estimating percent GFP expression {#s4a8} The transgenic marker that we used, rol-6(su1006), which confers a Roller phenotype ([@pbio-0020352-Kramer1]), does not allow us to identify transgenic embryos. Therefore, we estimated the fraction of transgenic embryos that express GFP as follows. Embryos from transgenic adults were collected and split into two samples. The first sample was scored for stage and GFP expression, while the second sample was allowed to develop and eventually scored for percent Roller animals. The percent Roller score was used as an estimate of the percent of animals that were transgenics. The percent transgenics with GFP expression was therefore estimated to be equal to (number of embryos expressing GFP) / ((total number of embryos scored) × (percent Roller)). Where reported in the text, numbers of transgenic embryos scored were estimated by this same approach. Supporting Information {#s5} ====================== Dataset S1 ::: {.caption} ###### Position Weight Matrices from Improbizer (14 KB PDF). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S1 ::: {.caption} ###### The 339 Microarray Positives (57 KB PDF). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S2 ::: {.caption} ###### List of 37 Ph-E Genes and 34 Ph-L Genes (14 KB PDF). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S3 ::: {.caption} ###### Conservation of Non-Coding Sequences in Ph-E and Ph-L Genes (22 KB PDF). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S4 ::: {.caption} ###### Occurrence of Conserved Temporal Elements and Predicted PHA-4 Sites in Ph-E and Ph-L Promoter Regions (18 KB PDF). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S5 ::: {.caption} ###### List of Genes Containing Conserved Sites of Different Motifs Within 500 bp Upstream of Their Predicted Start Codons (30 KB PDF). ::: ::: {.caption} ###### Click here for additional data file. ::: Accession Numbers {#s5a7} ----------------- The LocusLink (<http://www.ncbi.nlm.nih.gov/LocusLink>) accession numbers for the genes and gene products discussed in this paper are act-1 (LocusID 179535), C49G7.4 (LocusID 178809), ceh-22/Nkx 2--5 (LocusID 179485), Ce-pax-1 promoter (LocusID 187105), D-pax2 (LocusID 43825), end-1 (LocusID 179893), F30H5.3 (LocusID 175207), F47D12.7 (LocusID 175891), K07C11.4 (LocusID 179198), LAG-1 (LocusID 177373), M05B5.2 (LocusID 187451), myo-2 (LocusID 181404), ncr-1 (LocusID 180719), par-1 (LocusID 179912), pha-4 (LocusID 180357), R07B1.9 (LocusID 181201), rol-6(su1006) (LocusID 174397), skn-1 (LocusID 177343), T05E11.3 (LocusID 178014), vestigial (LocusID 36421), and Drosophila Su(H) (LocusID 34881). We are indebted to Stuart Kim and his laboratory for performing the microarray hybridizations, to Ken Kemphues and Jim Priess for sharing the unpublished KK822 par-1(zu310) strain, and to Jim Kent for the Improbizer program prior to publication. Thanks to Hiram Clawson for some of the C. elegans--C. briggsae alignments, available at <http://genome-test.cse.ucsc.edu/Intronerator/>, to Shale Dames for construction of some of the reporter constructs, and to Jeff Stevenson for assistance with the enhancer assays. JG was funded in part by a postdoctoral fellowship from the Canadian Institutes of Health Research. SEM is supported by R01 GM056264 and is an associate investigator of the Huntsman Cancer Institute. Additional funding for core facilities was provided by the National Institutes of Health 2P30CA42014. Conflicts of interest. The authors have declared that no conflicts of interest exist. Author contributions. JG and SEM conceived and designed the experiments. JG and MH performed the experiments. JG, SM, and SEM analyzed the data. JG and SEM wrote the paper. Academic Editor: Ronald H. A. Plasterk, Utrecht University ¤ Current address: Genes and Development Research Group, University of Calgary, Calgary, Alberta, Canada Citation: Gaudet J, Muttumu S, Horner M, Mango SE (2004) Whole-genome analysis of temporal gene expression during foregut development. PLoS Biol 2(11): e352. C : cluster score threshold (Cluster-Buster setting) Fox : forkhead box g : gap parameter (Cluster-Buster setting) GFP : green fluorescent protein m : motif score threshold (Cluster-Buster setting) Ph-E : set of early-expressed pharyngeal genes Ph-L : set of late-expressed pharyngeal genes PWM : position weight matrix	PubMed Central	2024-06-05T03:55:48.553276	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523228/", "journal": "PLoS Biol. 2004 Nov 19; 2(11):e352", "authors": [ { "first": "Jeb", "last": "Gaudet" }, { "first": "Srikanth", "last": "Muttumu" }, { "first": "Michael", "last": "Horner" }, { "first": "Susan E", "last": "Mango" } ] }
PMC523229	Introduction {#s1} ============ Thin layers of articular cartilage line the bones of synovial joints and provide a smooth, wear-resistant structure that reduces friction and absorbs impact forces ([@pbio-0020355-Brandt1]). Loss or damage to articular cartilage is a hallmark of arthritic diseases and is one of the most common reasons that both young and old adults seek medical care. Millions of people are afflicted with arthritis, and it ultimately affects more than half of people over the age of 65 ([@pbio-0020355-Badley1]; [@pbio-0020355-Yelin1]). A better understanding of the molecular mechanisms that create and maintain articular cartilage is crucial for discovering the causes of joint disorders and providing useful medical treatments. Joint formation begins during embryogenesis, when stripes of high cell density called interzones form across developing skeletal precursors ([@pbio-0020355-Haines1]). Programmed cell death occurs within the interzone, and a three-layered interzone forms that has two layers of higher cell density flanking a region of lower cell density. Non-joint precursors of the skeleton typically develop into cartilage, which hypertrophies and is replaced by bone. However, cells within the high-density layers of the interzone are excluded from this process and develop into the permanent layers of articular cartilage found in the mature joint ([@pbio-0020355-Mitrovic1]). Studies over the last 10 y have begun to elucidate some of the signaling pathways that contribute to the early stages of joint formation. Wnt14 is expressed in stripes at the sites where joints will form, and it is capable of inducing expression of other joint markers when misexpressed at new locations in the limb ([@pbio-0020355-Hartmann1]). Several members of the bone morphogenetic protein (BMP) family of secreted signaling molecules are also expressed in stripes at sites where joints will form, including those encoded by the genes Gdf5, Gdf6, Gdf7, Bmp2, and Bmp4 ([@pbio-0020355-Storm1]; [@pbio-0020355-Wolfman1]; [@pbio-0020355-Francis-West1]; [@pbio-0020355-Settle1]). Of these, Gdf5 expression is most strikingly limited to regions where joints will develop and is one of the earliest known markers of joint formation. Mutations in either Gdf5 or the closely related Gdf6 gene also block formation of joints at specific locations, providing strong evidence that these molecules are essential for the joint formation process ([@pbio-0020355-Storm2]; [@pbio-0020355-Settle1]). However, mutations in Bmp2 or Bmp4 cause early embryonic lethality, making it difficult to test their role in joint formation ([@pbio-0020355-Winnier1]; [@pbio-0020355-Zhang1]). Much less is known about how signaling pathways function during the subsequent maturation and maintenance of adult joint structures. Importantly, BMP signaling components are present in adult articular cartilage, suggesting that they may function during the late development or maintenance of this critical structure ([@pbio-0020355-Erlacher1]; [@pbio-0020355-Chubinskaya2]; [@pbio-0020355-Muehleman1]; [@pbio-0020355-Bau1]; [@pbio-0020355-Bobacz2]). BMPs bind tetrameric complexes of two type I and two type II transmembrane serine-threonine kinase receptors. Upon BMP binding, these complexes transduce a signal by phosphorylating members of the Smad family of transcription factors ([@pbio-0020355-Massague1]). Recent experiments have implicated two different BMP type I receptors in skeletal patterning, BMPR1A and BMPR1B. Both receptors can bind BMP2, BMP4, and GDF5, although GDF5 shows higher affinity for BMPR1B ([@pbio-0020355-Koenig1]; [@pbio-0020355-ten1]; [@pbio-0020355-Yamaji1]; [@pbio-0020355-Nishitoh1]; [@pbio-0020355-Chalaux1]). Both receptors are also expressed in dynamic patterns during normal development. In limbs, Bmpr1a expression becomes restricted to joint interzones, perichondrium, periarticular cartilage, hypertrophic chondrocytes, and interdigital limb mesenchyme. In comparison, Bmpr1b expression is seen primarily in condensing precartilaginous mesenchymal cells, regions flanking joint interzones, perichondrium, and periarticular cartilage ([@pbio-0020355-Dewulf1]; [@pbio-0020355-Mishina2]; [@pbio-0020355-Zou2]; [@pbio-0020355-Baur1]). Null mutations in the Bmpr1b gene produce viable mice with defects in bone and joint formation that closely resemble those seen in mice missing Gdf5 ([@pbio-0020355-Storm1]; [@pbio-0020355-Baur1]; [@pbio-0020355-Yi1]). Null mutations in Bmpr1a cause early embryonic lethality, with defects in gastrulation similar to those seen in mice with mutations in Bmp4 ([@pbio-0020355-Mishina2]; [@pbio-0020355-Winnier1]). Recent studies with floxed alleles suggest that Bmpr1a is also required for many later developmental events, but its roles in bone and joint formation have not yet been tested ([@pbio-0020355-Mishina1]). A genetic system for activating or inactivating genes specifically in joint tissues would be particularly useful for further studies of joint formation and maintenance. Here we take advantage of the tissue-specific expression pattern of the Gdf5 gene to engineer a Cre/loxP system ([@pbio-0020355-Nagy1]), Gdf5-Cre, that can be used to remove or ectopically express genes in joints. Tests with reporter mice show that this system is capable of modifying genes in all of the structures of the mature synovial joint, including the ligaments of the joint capsule, the synovial membrane, and the articular cartilage. Gdf5-Cre recombination bypasses the early embryonic lethality of null mutations in Bmpr1a, and shows that this receptor is required for early joint formation at some locations and for initiation of programmed cell death in webbing between digits. Interestingly, Bmpr1a is also required for postnatal maintenance of articular cartilage throughout most of the skeleton. In Gdf5-Cre/Bmpr1a^floxP^ mice, articular cartilage initially forms normally, but subsequently loses expression of several key cartilage markers after birth. It ultimately fibrillates and degenerates, resulting in severe osteoarthritis and loss of mobility. These experiments suggest that BMP signaling is required for normal maintenance of postnatal articular cartilage, and that modulation of the BMP signaling pathway may play an important role in joint disease. Results {#s2} ======= Genetic System for Testing the Function of Genes in Joint Development {#s2a} --------------------------------------------------------------------- To generate a general system capable of specifically testing genes for functions in skeletal joint development, we engineered transgenic mice to express Cre recombinase in developing joints ([Figure 1](#pbio-0020355-g001){ref-type="fig"}). Gdf5 is a gene strongly expressed in stripes across developing skeletal elements during embryonic joint formation. A bacterial artificial chromosome (BAC) containing the Gdf5 locus was modified by homologous recombination in bacteria to insert a cassette encoding Cre-internal ribosome entry site (IRES)-human placental alkaline phosphatase (hPLAP) into the translation start site of Gdf5 ([Figure 1](#pbio-0020355-g001){ref-type="fig"}A). This modified BAC was then used to make lines of transgenic mice. The resulting Gdf5-Cre transgenic mice were tested for transgene expression and Cre recombinase activity by crossing them to R26R reporter mice that activate the expression of lacZ after Cre-mediated removal of transcriptional stop sequences ([@pbio-0020355-Soriano1]). The resulting progeny were analyzed both for expression of the transgene by assaying HPLAP activity and for recombination of DNA by assaying LACZ activity. The progeny from all three lines showed strong LACZ expression primarily in joints, and in two of three lines HPLAP expression could also be seen in joint regions. Interestingly, HPLAP expression in the Gdf5-Cre transgenic GAC(A) line used for all subsequent breeding experiments was seen to precede LACZ expression during successive development of joints in the digits ([Figure 1](#pbio-0020355-g001){ref-type="fig"}C) (unpublished data). These experiments clearly demonstrate that the Gdf5-Cre transgene expresses Cre recombinase and causes DNA recombination in developing joint regions. ::: {#pbio-0020355-g001 .fig} Figure 1 ::: {.caption} ###### A Genetic System to Drive Gene Recombination in Developing Joints \(A) A 140-kb BAC from the Gdf5 locus was modified by inserting Cre-IRES-hPLAP into the translation start site of Gdf5 and used to make transgenic mice. Not to scale. See [Materials and Methods](#s4){ref-type="sec"} for details. (B--E) Visualization of Gdf5-Cre driven recombination patterns based on activation of lacZ expression from the R26R Cre reporter allele. (B) LACZ activity is visible as blue staining in the ear (ea) and the joints of the shoulder (s), elbow (eb), wrist (w), knee (k), ankle (a), vertebra (vj), and phalanges (black arrowheads) of an E14.5 mouse embryo. (C) E14.5 hindlimb double-stained to show both HPLAP expression from the transgene (grey/purple staining) and LACZ expression from the rearranged R26R allele (blue staining). Note that both markers are visible in the oldest, proximal interphalangeal joint (black arrowhead), only HPLAP activity is visible in the more recently formed medial interphalangeal joint (black arrow), and neither HPLAP nor LACZ expression is visible in the youngest, most distal joint of the digit (white arrowhead). (D) Newborn (P0) forelimb with skin partially removed showing LACZ activity expressed in all phalangeal joints (red Salmon gal staining, black arrowheads) and regions of some tendons (asterisk). (E) Section through the most distal phalangeal joint of a P0 hindlimb stained with Alcian blue to mark cartilage showing LACZ expression (stained red) in all tissues of developing joints: articular cartilage (black arrowhead), precursors of ligaments and synovial membranes (black arrow), and cells where cavitation is occurring (asterisk). ::: ![](pbio.0020355.g001) ::: GAC(A) mice were crossed with lacZ ROSA26 Cre reporter strain (R26R) mice to analyze the pattern of Cre-mediated lacZ recombination throughout development. Joints in developing limbs begin forming in a proximal-distal pattern such that the shoulder joint forms prior to the elbow joint. In addition, three major stages of early joint development have been defined by histology as (1) interzone formation, (2) three-layer interzone formation, and (3) cavitation ([@pbio-0020355-Mitrovic1]). Consistent with the proximal-distal pattern of joint development in the limbs, LACZ activity is seen at embryonic day 12.5 (E12.5) in the more proximal joints, including the shoulder and knee (unpublished data). By E14.5, LACZ expression is typically seen in all but the most distal joints of the limbs ([Figure 1](#pbio-0020355-g001){ref-type="fig"}B and [1](#pbio-0020355-g001){ref-type="fig"}C), but with some variability in both strength and extent of expression from embryo to embryo. The strongest-staining embryos often have additional staining in fingertips (not seen in the E14.5 embryo in [Figure 1](#pbio-0020355-g001){ref-type="fig"}C, but clearly detectable in the E13.5 embryo shown in [Figure 2](#pbio-0020355-g002){ref-type="fig"}). Sections through developing joints show that LACZ is present in many cells at the interzone stage (unpublished data). However, expression of LACZ in nearly 100% of joint cells is not achieved until the three-layer interzone stage (for example, in the knee joint at E14.5 or in any of the phalangeal joints at E16.5 (unpublished data). Within the developing skeleton, Cre-mediated expression of LACZ remains strikingly specific to joints throughout development. Furthermore, it is seen in all the structures of postnatal synovial joints including the articular cartilage, joint capsule, and synovial membrane ([Figure 1](#pbio-0020355-g001){ref-type="fig"}D and [1](#pbio-0020355-g001){ref-type="fig"}E) (unpublished data). These patterns are consistent with the well-established expression of Gdf5 in interzone regions during embryonic development ([@pbio-0020355-Storm1]). Adult expression patterns of the Gdf5 gene are not as well characterized, but Gdf5 expression has previously been detected in adult articular cartilage using both RT-PCR and immunocytochemistry ([@pbio-0020355-Chang1]; [@pbio-0020355-Erlacher1]; [@pbio-0020355-Bobacz1]). ::: {#pbio-0020355-g002 .fig} Figure 2 ::: {.caption} ###### Bmpr1a Is Required for Webbing Regression and Apoptosis in Specific Regions of the Limb (A and B) Control E14.5 forelimb (A) compared to a, E14.5 mutant forelimb (B) showing webbing between digits 1 and 2 (arrowheads) and extra tissue at the posterior of digit 5 (arrows). \(C) Gdf5-Cre induced lacZ expression from R26R in an E13.5 forelimb showing LACZ staining (blue) in metacarpal-phalangeal joints, between digits 1 and 2 (arrowhead), and in a region posterior to digit 5 (arrow). (D and E) Sections of E14.5 hindlimbs showing apoptosis visualized by TUNEL staining (green) and proliferation visualized by staining for histone H3 phosphorylation (red). Controls show strong, uniform TUNEL staining between digits 1 and 2 (D, arrowhead) while mutants show patchy TUNEL staining interspersed with mitotic cells in similar regions (E). Scale bar = 200 μm. \(F) Quantitation of TUNEL staining and mitotic cells in the posterior region of the fifth digit shows apoptosis is reduced 30% while proliferation is increased 20% (asterisks indicate statistically significant difference). (G and H) By E15.5, interdigital tissue has regressed in controls (G, arrowhead). In contrast, tissue remains in mutants at this location, primarily derived from cells that have undergone Gdf5-Cre-mediated recombination that inactivates Bmpr1a function and activates expression of LACZ (H). Scale bar = 75 μm. ::: ![](pbio.0020355.g002) ::: Other sites besides limb joints also have Cre-mediated lacZ expression. Starting at E13.5, LACZ activity is detected in an anterior and posterior domain of the limb bud ([Figure 2](#pbio-0020355-g002){ref-type="fig"}C). At E14.5, LACZ activity is detectable in the developing ear pinnae, ribs, sternum, tissues in the face, and some regions of the brain and spinal cord ([Figure 1](#pbio-0020355-g001){ref-type="fig"}B) (unpublished data). At birth, LACZ is also expressed in tendons running along the vertebral column, regions of tendons in the wrist and ankle, and some tendon insertions ([Figure 1](#pbio-0020355-g001){ref-type="fig"}D) (unpublished data). By 5 wk of age, LACZ is also expressed in the hair follicles, ear cartilage, some cells in the growth plate of the long bones, and portions of the brain and spinal cord (unpublished data). Surprisingly, 23 of 63, or 37% of transgenic mice analyzed also show some degree of wider "ectopic" LACZ expression, which can extend throughout many different tissues in the animal. However, sustained expression of the transgene itself, as assayed by HPLAP activity, is still restricted primarily to joints in animals that show evidence of more generalized recombination based on LACZ expression (unpublished data). This suggests that in a fraction of animals, sporadic expression of Cre at some time early in development is sufficient to lead to both ectopic recombination and LACZ expression. While the fraction of animals with broader recombination patterns must be tracked and accounted for during experiments, these animals offer the potential benefit of revealing additional new functions of target genes that could be subsequently studied with additional site-specific Cre drivers. Gdf5-Cre/Bmpr1a^floxP^ Animals Survive to Adulthood with Ear, Webbing, and Joint Defects {#s2b} ------------------------------------------------------------------------------------------ We next used the Gdf5-Cre system to test the role of BMP signaling during normal joint development. Gdf5-Cre transgenic mice were bred to animals carrying a conditional floxed allele of the Bmpr1a locus ([@pbio-0020355-Mishina3]), usually in the presence of the R26R reporter allele to facilitate simultaneous visualization of Cre-mediated recombination patterns (see typical cross in [Figure 3](#pbio-0020355-g003){ref-type="fig"}). PCR amplification confirmed that a key exon of the Bmpr1a gene was deleted in mice that also carried the Gdf5-Cre transgene (unpublished data). Previous studies have shown that the recombined Bmpr1a^floxP^ allele mimics a null allele of the Bmpr1a locus when transmitted through the germline ([@pbio-0020355-Mishina3]). The Gdf5-Cre/Bmpr1a^floxP^ conditional knockout mice were viable and survived to adulthood, showing that the Gdf5-Cre driver can bypass the early embryonic lethality previously reported in animals with a null mutation in the Bmpr1a locus ([@pbio-0020355-Mishina2]). ::: {#pbio-0020355-g003 .fig} Figure 3 ::: {.caption} ###### Gdf5-Cre-Mediated Deletion of Bmpr1a \(A) Breeding strategy simultaneously deletes Bmpr1a^floxP^ and allows visualization of Gdf5-Cre-mediated recombination by lacZ expression from R26R. (B--E) 5-week-old mutant and control mice stained with Alcian blue to mark cartilage and alizarin red to mark bone. (B) Ankle of control with strong blue staining lining each joint (arrowheads). (C) Ankle of mutant showing an absence of blue staining in most regions (arrowheads) and a joint fusion between the central (c) and second (2) tarsals (arrow). (D) Control and (E) mutant metatarsal/phalangeal joint which lacks blue staining in articular regions (arrowheads) but retains staining in the growth plate (asterisks). \(F) Control forelimb. \(G) Mutant forelimb with webbing between the first and second digit (black arrowhead). ::: ![](pbio.0020355.g003) ::: The viable Gdf5-Cre/Bmpr1a^floxP^ mice showed several phenotypes. First, the conditional knockout mice had shorter ears that often lay flatter against their heads than controls (controls 13.1 ± 0.1 mm long, n = 38; mutants 11.8 ± 0.2 mm, n = 11; p \< 0.0001). BMP signaling is known to be required for growth of the external ear of mice ([@pbio-0020355-Kingsley1]), and this phenotype likely reflects loss of Bmpr1a function in the fraction of ear cells that express the Gdf5-Cre transgene. Most mutant mice also showed soft tissue syndactyly or retention of webbing between the first and second digits of their feet, a phenotype that was more frequent and more severe in the forelimbs (201 of 220, or 91%, of forefeet and 109 of 220, or 50%, of hindfeet). Finally, mutant animals showed obvious skeletal changes in whole-mount skeletal preparations. At some sites in the ankles, joints seemed to be missing entirely, with fusion of bones that would normally be separate. For example, the second distal tarsal was fused to the central tarsal bone in every conditional knockout animal examined (18 of 18), a phenotype not observed in controls (zero of 18) ([Figure 3](#pbio-0020355-g003){ref-type="fig"}B and [3](#pbio-0020355-g003){ref-type="fig"}C). At other locations, joints had clearly formed but showed dramatic loss of staining with the cartilage matrix marker Alcian blue ([Figure 3](#pbio-0020355-g003){ref-type="fig"}B--[3](#pbio-0020355-g003){ref-type="fig"}E) (unpublished data). Normal Alcian blue staining was seen in non-articular regions, such as the cartilaginous growth plate ([Figure 3](#pbio-0020355-g003){ref-type="fig"}D and [3](#pbio-0020355-g003){ref-type="fig"}E, asterisk). These data suggest that Bmpr1a function is required for the formation of specific joints in the ankle region and for either generation or maintenance of articular cartilage in most other joints of the limb. Developmental Origin of Webbing Phenotype {#s2c} ----------------------------------------- Interdigital mesenchyme is normally eliminated by apoptosis during embryonic development, a process that can be stimulated by BMP beads, inhibited by Noggin, or blocked by overexpression of dominant-negative BMP receptors ([@pbio-0020355-Garcia-Martinez1]; [@pbio-0020355-Yokouchi1]; [@pbio-0020355-Zou1]; [@pbio-0020355-Guha1]). Limbs of Gdf5-Cre/Bmpr1a^floxP^ mutant embryos showed obvious retention of interdigital webbing between the first and second, but not other, digits of E14.5 forelimbs ([Figure 2](#pbio-0020355-g002){ref-type="fig"}A and [2](#pbio-0020355-g002){ref-type="fig"}B), a pattern that corresponds to the presence or absence of webbing seen in the adult limb. They also showed excess tissue on the posterior margin of the fifth digit ([Figure 2](#pbio-0020355-g002){ref-type="fig"}B, arrow). Analysis of LACZ expression in Gdf5-Cre/R26R reporter embryos showed that Cre-mediated recombination has occurred by E13.5 in the metacarpal-phalangeal joints, and in the interdigital region between the first and second, but not other, digits. In addition, a domain of recombination and expression of LACZ is also reproducibly seen in the posterior half of the fifth digit ([Figure 2](#pbio-0020355-g002){ref-type="fig"}C). Terminal deoxynucleotidyl transferase--mediated deoxyuridine triphosphate nick end labeling (TUNEL) staining of interdigital mesenchyme between the first and second digits ([Figure 2](#pbio-0020355-g002){ref-type="fig"}D and [2](#pbio-0020355-g002){ref-type="fig"}E) and the fifth digit flanking mesenchyme showed a decreased number of dying cells in the regions where excess tissue is retained in the mutant limbs. Numbers of phosphorylated histone H3-labeled proliferating cells were also elevated in these regions ([Figure 2](#pbio-0020355-g002){ref-type="fig"}F). Most cells found in the webbed region between the first and second digits at E15.5 strongly expressed LACZ in Gdf5-Cre/Bmpr1a^floxP^ mutant embryos ([Figure 2](#pbio-0020355-g002){ref-type="fig"}H). These data suggest that regional loss of BMPR1A receptor signaling blocks programmed cell death in interdigital mesenchyme, and that the recombined cells survive and proliferate in the absence of BMPR1A signaling. Failure of Early Joint Formation in Ankle Regions {#s2d} ------------------------------------------------- The Bmpr1a gene is expressed in the interzone region of developing joints at E13.5 ([@pbio-0020355-Baur1]). In situ hybridization showed that the gene is also expressed in the interzones of ankle joints and prospective articular cartilage regions of digit joints at E15.5 ([Figure 4](#pbio-0020355-g004){ref-type="fig"}). LACZ staining indicated that Cre-mediated recombination begins to occur in ankle joints around E14.5, and is extensive by E15.5 ([Figure 4](#pbio-0020355-g004){ref-type="fig"}G and [4](#pbio-0020355-g004){ref-type="fig"}J) (unpublished data). In the ankle joint regions that were obviously fused in postnatal mutant animals, alterations in early joint marker expression could also be seen by E15.5. At this stage, the Gdf5 gene is normally expressed in stripes that mark the sites of joint formation ([Figure 4](#pbio-0020355-g004){ref-type="fig"}F), and the gene for the major collagen protein of cartilage matrix (Col2a1) is down-regulated in the interzone region ([Figure 4](#pbio-0020355-g004){ref-type="fig"}E). In contrast, Col2a1 staining extended completely through the joint region between the second and central tarsal of Gdf5-Cre/Bmpr1a^floxP^ mutants ([Figure 4](#pbio-0020355-g004){ref-type="fig"}H, black arrow), and Gdf5 expression was seen only as a small notch extending into where the joint should be forming ([Figure 4](#pbio-0020355-g004){ref-type="fig"}I, bracket). These data suggest that the fusions seen between ankle bones in postnatal mutant skeletons are the result of incomplete segmentation of skeletal precursors during embryonic development, a defect confined to some locations in the ankle. ::: {#pbio-0020355-g004 .fig} Figure 4 ::: {.caption} ###### Bmpr1a Is Expressed in Joints and Is Required for Continued Joint Formation in the Ankle Region \(A) Diagram of ankle bones from a wild-type mouse; bones fusing in mutant are colored red. Roman numerals II--IV, metatarsals; 2, 3, and 4/5, distal row of tarsal bones; c, central tarsal bone; ta, talus; ca, calcaneus. (B and C) In situ hybridization at E15.5 showing that Bmpr1a is expressed in ankle joint interzones (B, arrowheads) and in the forming articular regions of the phalangeal joints (C, arrowheads). \(D) Near adjacent section to (C) showing Gdf5-Cre induced LACZ expression from R26R in the forming joints of the digits (arrowheads). (E--J) Marker gene expression and R26R LACZ staining patterns on near adjacent sections of control and mutant embryos. In control mice at E15.5 ankle joints are clearly delineated as regions that have down-regulated Col2 (E), express Gdf5 throughout (F), and express LACZ in most cells (G; white arrowheads and black arrows). In mutant embryos at the same stage, joint formation is incomplete. Faint Col2 expression can be seen connecting a medial region of tarsal 2 with metatarsal II (H, white arrowhead), and Gdf5 expression does not extend all the way across the joint at this location (I, white arrowhead). Between tarsals c and 2, mutants express Col2 across the normal joint-forming region (H, black arrow) and lack expression of Gdf5 at sites where skeletal fusions are observed (I, black arrow and bracket). (J) Scale bar = 100 μm. ::: ![](pbio.0020355.g004) ::: Failure to Maintain Articular Cartilage in Other Joints {#s2e} ------------------------------------------------------- In most joints of Bmpr1a conditional knockout mice, embryonic segmentation of skeletal precursors occurred normally. Although Gdf5-Cre-mediated recombination was seen as early as E13.5 in digit interzone regions (see [Figure 2](#pbio-0020355-g002){ref-type="fig"}C), no changes in cell death or cell proliferation could be seen in the metacarpal-phalangeal or metatarsal-phalangeal joints at E13.5 or E14.5 (unpublished data). Similarly, although clear LACZ expression was seen by E15.5 in interphalangeal joints and periarticular regions ([Figure 4](#pbio-0020355-g004){ref-type="fig"}D), no difference in morphology or expression of Col2a1, Gdf5, or Bmpr1b was seen in the articular regions of the phalanges at these stages (unpublished data). At birth, digit joints were generally indistinguishable from those in control animals; chondrocytes were abundant in articular regions and were surrounded by typical cartilage matrix with normal staining by Safranin O, a histological stain for proteoglycans ([Figure 5](#pbio-0020355-g005){ref-type="fig"}). At this stage, both wild-type and mutant cells in articular regions also expressed high levels of Col2a1 and Aggrecan (Agg), the genes encoding the major structural proteins of cartilage matrix ([Figure 5](#pbio-0020355-g005){ref-type="fig"}B and [5](#pbio-0020355-g005){ref-type="fig"}G) (unpublished data). No alterations in cellular apoptosis or proliferation were observed (unpublished data). ::: {#pbio-0020355-g005 .fig} Figure 5 ::: {.caption} ###### Bmpr1a Is Required to Maintain Expression of ECM Components in Articular Cartilage In situ hybridization or LACZ staining on near adjacent sections of metacarpal-phalangeal joints (A--C and F--H) and the tarsal 2-metatarsal II joint (D--E and I--J) of P0 mice. At birth, articular cartilage of controls (A--E) and mutants (F--J) appears similar by Safranin O staining (A and F), and Col2 expression (B, G). Mat4 expression confirms that articular cartilage is initially specified in mutants (D andI, brackets). LACZ expression confirms Cre-mediated recombination has occurred in articular cartilage (C, H, E, and J). (K--T) Near adjacent sections of the metacarpal-phalangeal joints of P14 mice. Two weeks after birth, articular cartilage of controls stains with pericellular Safranin O (orange staining, K), and expresses Col2 (L), Agg (M), and SOX9 (N). In contrast, mutant articular cells are smaller and more densely packed, lack pericellular Safranin O staining (P), have reduced expression of Col2 (Q) and Agg (R), but retain normal levels of SOX9 protein (S, brackets; dashed line marks faint edges of articular surfaces). LACZ expression confirms Cre-mediated recombination has occurred in articular cells (O ansd T, brackets). (A and K) Scale bar = 75 μm. ::: ![](pbio.0020355.g005) ::: To determine whether articular cells were properly specified in mutants, we also analyzed expression of Matrilin-4 (Mat4), a gene expressed specifically in the periarticular and perichondral regions of developing joints ([@pbio-0020355-Klatt1]). In both control and mutant animals, transcription of Mat4 was clearly detectable in the articular cartilage layers of newborn joints ([Figure 5](#pbio-0020355-g005){ref-type="fig"}D and [5](#pbio-0020355-g005){ref-type="fig"}I). In all experiments, expression of LACZ throughout articular regions indicated that Cre-mediated recombination had occurred throughout the articular regions ([Figure 5](#pbio-0020355-g005){ref-type="fig"}C, [5](#pbio-0020355-g005){ref-type="fig"}H, [5](#pbio-0020355-g005){ref-type="fig"}E, and [5](#pbio-0020355-g005){ref-type="fig"}J). The normal histological appearance, staining properties, and marker gene expression patterns suggest that Bmpr1a is not required for the initial formation or specification of articular cartilage. By 1 wk after birth, obvious differences began to be detected in the articular regions of mutant animals. The expression of Col2a1 was reduced throughout the articular surfaces of the carpals, metacarpals, and phalanges of the forefeet (unpublished data). Less severe reductions were also seen in articular cells of tarsals and metatarsals in the hindfeet (unpublished data). By 2 wk of age, Col2a1 expression was reduced in most cells of the articular region ([Figure 5](#pbio-0020355-g005){ref-type="fig"}L and [5](#pbio-0020355-g005){ref-type="fig"}Q), accompanied by markedly reduced Safranin O staining ([Figure 5](#pbio-0020355-g005){ref-type="fig"}K and [5](#pbio-0020355-g005){ref-type="fig"}P), and decreased expression of Agg and two genes normally expressed in more mature articular cartilage cells, Collagen 3 (Col3a1) and Collagen 10 (Col10a1) ([Figure 5](#pbio-0020355-g005){ref-type="fig"}M and [5](#pbio-0020355-g005){ref-type="fig"}R) (unpublished data) ([@pbio-0020355-Eyre1]). Inhibition of BMP signaling in cultured chondrocytes has previously been reported to induce Collagen 1 (Col1a1) expression, increase proliferation, and result in cells with flattened, fibroblast-like morphology ([@pbio-0020355-Enomoto-Iwamoto1]). However, we saw no increase in the expression of Col1a1 in mutant articular cartilage, and no proliferation was detected in articular cells of either mutant or control animals (unpublished data). While recombined LACZ marker expression was detected in most articular cartilage cells, it was also observed in scattered subarticular chondrocytes, growth plate chondrocytes, and osteoblasts ([Figure 5](#pbio-0020355-g005){ref-type="fig"}O and [5](#pbio-0020355-g005){ref-type="fig"}T) (unpublished data). Although this implies that BMP signaling was defective in multiple cell types, the observed defects were confined to the articular cartilage. For example, Osteocalcin and Col1a1 expression appeared normal in osteoblasts (unpublished data). Together, these data suggest that BMPR1A activity is required in postnatal joint articular cartilage to maintain expression of many genes encoding structural components of cartilage matrix. Previous studies have shown that Sox9 is required for normal cartilage differentiation, for expression of cartilage extracellular matrix (ECM) genes including Agg, and is a direct transcriptional regulator of the key cartilage matrix gene Col2a1 ([@pbio-0020355-Bell1]; [@pbio-0020355-Lefebvre2]; [@pbio-0020355-Bi1]; [@pbio-0020355-Sekiya1]). Notably, despite reduced expression of many cartilage matrix marker genes in Bmpr1a mutant mice, the SOX9 protein was present at normal levels in articular regions at all stages examined, including newborn, 2-wk-old, 7-wk-old, and 9-mo-old mice ([Figure 5](#pbio-0020355-g005){ref-type="fig"}N and [5](#pbio-0020355-g005){ref-type="fig"}S) (unpublished data). Synovial Hypertrophy, Cartilage Erosion, and Accelerated Cartilage Maturation {#s2f} ----------------------------------------------------------------------------- Conditional loss of Bmpr1a led to marked hypertrophy of the synovial membrane in the joint capsule of some joints, particularly in the ankle region. In the most severely affected joints, the expanded synovial membrane grew into the joint space and was associated with obvious loss or erosion of the articular cartilage ([Figure 6](#pbio-0020355-g006){ref-type="fig"}A and [6](#pbio-0020355-g006){ref-type="fig"}B, asterisks, arrows). Accelerated cartilage maturation and increased expression of Col10a1 was frequently seen in the chondrocytes underlying the articular erosions ([Figure 6](#pbio-0020355-g006){ref-type="fig"}C and [6](#pbio-0020355-g006){ref-type="fig"}D, brackets) (unpublished data). Interestingly, the regions of increased Col10a1 expression did not correspond to the regions that had undergone Cre-mediated recombination. Instead, increased expression of Col10a1 was seen in a zone of largely LACZ-negative cells stretching from the cartilage adjacent to the ossification front (where Col10a1 is normally expressed in maturing cartilage cells), toward the regions where surface articular cartilage was severely eroded or missing ([Figure 6](#pbio-0020355-g006){ref-type="fig"}A and [6](#pbio-0020355-g006){ref-type="fig"}B, arrowheads). Previous studies suggest that parathyroid hormone-related protein, a diffusible signal made in the articular surface, may normally inhibit maturation of underlying cartilage ([@pbio-0020355-Vortkamp1]; [@pbio-0020355-Weir1]). Local loss of the articular surface could remove this inhibition and lead to a cell-nonautonomous acceleration of maturation in chondrocytes underlying points of articular erosion. ::: {#pbio-0020355-g006 .fig} Figure 6 ::: {.caption} ###### Synovial Membrane Expansion, Articular Surface Erosion, and Accelerated Maturation of Underlying Cartilage in Ankles of Bmpr1a Mutant Mice Near adjacent sections from the tarsal 2-metatarsal II joint of 7-d-old mice. (A and B) LACZ staining (blue) shows Cre-mediated recombination is largely restricted to articular (arrowheads) and synovial cells (asterisks) in both controls and mutants. (C and D) In situ hybridization shows Col10 expression expands in mutants toward regions of synovial membrane expansion and articular surface erosion (brackets and arrows). This may be a cell nonautonomous effect of joint damage, since the LACZ expressing cells at the articular surface do not show upregulation of Col10 (arrowheads) and the region of expanded Col10 expression is largely made up of cells that have not undergone Cre-mediated recombination. Note the formation of a cartilaginous bridge along the joint capsule of the mutant where joint formation is disrupted at earlier stages (B, white arrowhead, and [Figure 3](#pbio-0020355-g003){ref-type="fig"}, white arrowheads). (A) Scale bar = 75 μm. ::: ![](pbio.0020355.g006) ::: This synovial hypertrophy is associated with increased numbers of mononuclear cells resembling synoviocytes or macrophages, cell types that are difficult to distinguish even with surface markers at early postnatal stages. However, no neutrophils were observed, suggesting that there is little inflammation. At later stages synovial hypertrophy is reduced. Further work will be needed to determine whether synovial development is regulated by BMP signaling, or whether the synovium becomes enlarged as a response to nearby skeletal malformations (such as fusion of the second and central tarsals or defects in the articular cartilage). Noninflammatory Degeneration of Articular Cartilage in Digit and Knee Joints {#s2g} ---------------------------------------------------------------------------- Outside of the ankle region, little or no evidence was seen for expansion of the synovial membrane. Instead, mutant mice showed histological signs of osteoarthritis, such as fibrillation of the articular surface ([Figure 7](#pbio-0020355-g007){ref-type="fig"}). As previously seen in 1- and 2-wk-old animals, Safranin O staining and Agg and Col10 expression were all reduced in mutant articular regions of the forefeet and hindfeet by 7 wk of age, and the beginning signs of cartilage loss were observed (unpublished data). By 9 mo of age, many regions of articular cartilage were completely missing or extremely fibrillated, leaving regions of exposed bone on the surface ([Figure 7](#pbio-0020355-g007){ref-type="fig"}A--[7](#pbio-0020355-g007){ref-type="fig"}D). No alterations were seen in the expression of Osteocalcin, Col1a1, or matrix metalloprotease-13 at either 7 wk or 9 mo. ::: {#pbio-0020355-g007 .fig} Figure 7 ::: {.caption} ###### Loss of Bmpr1a Signaling Leads to Articular Cartilage Fibrillation and Degeneration in Digits and Knees of Aging Mice (A--D) Near adjacent sections of metatarsal-phalangeal joints from 9 month old mice. Articular cartilage of controls is complete and stains strongly with Safranin O (A, orange stain). In contrast, articular cells of mutants are severely fibrillated or absent with much reduced staining of Safranin O (C, arrowheads). LACZ expression confirms Cre-mediated recombination has occurred in articular cells (B and D). (E--P) Sagittal sections through knee joints of 7-wk- (E--J) or 9-mo-old animals (K--P); fe, femur; ti, tibia; gp, growth plate. Seven weeks after birth, the height of the tibial epiphysis is reduced in mutants (E and H, bars), and their articular layer stains poorly with Safranin O, is fibrillated, and is strikingly thinner (F and I, black arrowhead, and brackets). Near adjacent sections with LACZ staining confirm Cre-mediated recombination has occurred in articular cells (G and J). Note that in mutants, LACZ is absent in cells adjacent to those that do stain with Safranin O, suggesting Bmpr1a may act cell autonomously (I and J, white arrowheads). At 9 mo old, the mutant tibial epiphysis is extremely thin (K and N, bars), and the articular layer is completely absent, leaving bone to rub directly on bone (L and O, bracket). LACZ staining shows Cre-mediated recombination occurred in articular cells of controls (M) and in some remaining skeletal tissue of mutants (P). Also note aberrantly formed meniscal cartilage in mutants (E, H, K, and N, arrows), and increased sclerosis in mutant epiphyses (E, H, K, and N, asterisks). (A and K) Scale bar = 50 μm; (I) scale bar = 300 μm. ::: ![](pbio.0020355.g007) ::: The major weight-bearing joint of the hindlimb, the knee, showed changes that closely paralleled that seen in the foot joints. All markers of cartilage matrix looked similar to controls at E16.5, suggesting that early stages of joint formation were not disrupted (unpublished data). By postnatal day 7, Safranin O staining and Col2a1 and Agg expression were clearly reduced in the mutant, despite continued expression of Sox9 (unpublished data). The overall shape of mutant knee skeletal elements appeared similar to controls, although the fibrocartilaginous meniscus that resides between the femur and tibia appeared much less dense in mutants at E16.5. Some cartilage formed in the meniscus region, but the size of these elements was greatly reduced and contained abundant cells with fibrous, noncartilaginous appearance (unpublished data). This reduction of the meniscus can also be seen in sections from 7-wk- and 9-mo-old animals ([Figure 7](#pbio-0020355-g007){ref-type="fig"}E, [7](#pbio-0020355-g007){ref-type="fig"}H, [7](#pbio-0020355-g007){ref-type="fig"}K, and [7](#pbio-0020355-g007){ref-type="fig"}N, arrows). At 7 wk of age the normally domed tibial epiphysis was flattened and depressed in the knees of mutant animals, markedly reducing the distance between the growth plate and articular surface ([Figure 7](#pbio-0020355-g007){ref-type="fig"}E and [7](#pbio-0020355-g007){ref-type="fig"}H, vertical bar). Articular cartilage was also thinner than in control animals, showed nearly complete absence of Safranin O staining, and was either acellular or beginning to fibrillate in many regions ([Figure 7](#pbio-0020355-g007){ref-type="fig"}F and [7](#pbio-0020355-g007){ref-type="fig"}I). The few large Safranin O-stained cells still apparent in mutant articular regions appeared to correspond in position to rare LACZ-negative cells in adjacent sections, suggesting that Bmpr1a is required cell-autonomously in articular cartilage ([Figure 7](#pbio-0020355-g007){ref-type="fig"}I and [7](#pbio-0020355-g007){ref-type="fig"}J, white arrowheads). By 9 mo, large areas of mutant knees were devoid of articular cells, and the bones of the femur and tibia appeared to rub directly against each other. Furthermore, the epiphysis of the tibia was extremely depressed, to the point that growth plate cartilage was almost exposed through the surface of the bone ([Figure 7](#pbio-0020355-g007){ref-type="fig"}K, [7](#pbio-0020355-g007){ref-type="fig"}L, [7](#pbio-0020355-g007){ref-type="fig"}N, and [7](#pbio-0020355-g007){ref-type="fig"}O). In addition, mutants at 7 wk and 9 mo showed subchondral sclerosis, especially in the epiphysis of the femur ([Figure 7](#pbio-0020355-g007){ref-type="fig"}E, [7](#pbio-0020355-g007){ref-type="fig"}H, [7](#pbio-0020355-g007){ref-type="fig"}K, and [7](#pbio-0020355-g007){ref-type="fig"}N, asterisks). While subchondral sclerosis is commonly seen in cases of osteoarthritis, it is unclear in this case whether the sclerosis is mainly a response of bone formation to compensate for decreased articular cartilage, or whether it is the effect of loss of Bmpr1a signaling in some LACZ-positive cells that are also observed in these regions (unpublished data). The histological signs of joint arthritis were accompanied by functional impairments in both grasping ability and range of motion in mutant animals. Gdf5-Cre/Bmpr1a^floxP^ mutant animals showed a highly significantly reduced ability to grasp and remain suspended on a slender rod (mean suspension time: controls 38 ± 6 s, n = 39; mutants 6 ± 3 s, n = 11; p \< 0.0001). Mutant mice also showed a clear decrease in the maximum range of mobility of two different joints in the digits, as assayed by passive manipulation (MT/P1 joint: controls 100 ± 0°, n = 26; mutants 82 ± 3°, n = 8; p \< 0.0003; P1/P2 joint: controls 152 ± 1°, n = 23; mutants 140 ± 5°, n = 6; p \< 0.05). The structural, histological, marker gene expression, and functional changes in mutant mice demonstrate that BMPR1A is required for normal postnatal maintenance of articular cartilage. Discussion {#s3} ========== Previous studies suggest that BMP signaling is involved in a large number of developmental events. Many of these events occur early in embryogenesis, and complete inactivation of BMP receptors causes death by E9.5 ([@pbio-0020355-Mishina2]). The Gdf5-Cre recombination system bypasses the early embryonic lethality of Bmpr1a mutations, and provides important new information about the role of this receptor in limb and skeletal development. The three major limb phenotypes revealed by eliminating Bmpr1a with Gdf5-driven Cre include webbing between digits, lack of joint formation at specific locations in the ankle, and failure to maintain articular cartilage after birth, resulting in severe arthritis. Previous studies have shown that manipulation of BMP signaling alters interdigital apoptosis during development of the limb, but no experiment has identified a specific member of the BMP signaling pathway that is required for this process ([@pbio-0020355-Yokouchi1]; [@pbio-0020355-Zou1]; [@pbio-0020355-Zou2]; [@pbio-0020355-Guha1]). Our new loss-of-function data confirm that BMP signaling is required for interdigital apoptosis and suggests that Bmpr1a is a critical component for mediating this signal. At some sites, loss of Bmpr1a function leads to a defect in the early stages of joint formation, resulting in a complete failure to form a joint and fusion of bones in the ankle. Mutations in two different ligands in the BMP family, Gdf5 and Gdf6, the Bmpr1b receptor, and in the human Noggin locus ([@pbio-0020355-Storm1]; [@pbio-0020355-Gong1]; [@pbio-0020355-Baur1]; [@pbio-0020355-Yi1]; [@pbio-0020355-Settle1]) also produce defects in joint formation at specific locations in the limbs. The joint defects associated with multiple components of the BMP pathway provide strong evidence that BMP signaling is required for early stages of joint formation at some anatomical locations. Most joints still form normally when Bmpr1a is knocked out in Gdf5 expression domains. The lack of joint fusions outside the ankle region could be due to differences in requirement for BMP signaling in different joints, to compensating expression of other BMP receptors outside the ankles, or to differences in the detailed timing of Gdf5-Cre stimulated gene inactivation in ankles and other joint regions. Comparison of the expression of the HPLAP marker (driven directly by Gdf5 control elements) and the R26R LACZ marker (expressed following Gdf5-Cre recombination) suggests that recombination-stimulated changes in gene expression may be delayed for a 0.5--1 d in the digit region (see [Figure 1](#pbio-0020355-g001){ref-type="fig"}C). In addition, levels of Bmpr1a mRNA and protein may persist for some time following Gdf5-Cre stimulated recombination, making it possible to bypass an early requirement for Bmpr1a in joint formation at some locations. Following the decay of Bmpr1a mRNA and protein, the Gdf5-Cre strategy should result in permanent inactivation of Bmpr1a function in recombined cells. This system thus provides one of the first strong genetic tests of Bmpr1a function at later stages of joint development. Despite the normal appearance of articular regions and gene expression immediately after birth, Bmpr1a-deficient animals are unable to maintain the normal differentiated state of articular cartilage as they continue to develop and age. These results suggest that BMP receptor signaling is essential for continued health and integrity of articular cartilage in the postnatal period. Articular cartilage is a key component of synovial joints and is one of the few regions in the skeleton where cartilage is maintained into adulthood. Despite the importance of articular cartilage in joint health and mobility, little is known about the factors that create and maintain it in thin layers at the ends of long bones. In our experiments, articular cartilage lacking Bmpr1a retains some normal characteristics, in that it maintains a very low proliferation rate, does not express Col1a1, and continues to express SOX9, a major transcription factor regulating expression of structural components of cartilage matrix. However, several of the most prominent structural components of cartilage matrix fail to be maintained in mutant animals, resulting in decreased synthesis of Col2a1, Agg, and proteoglycans. Therefore, BMPR1A appears to maintain articular cartilage primarily through inducing expression of key ECM components. It is interesting that the SOX9 transcription factor continues to be expressed in mutant cartilage despite loss of Col2a1, a direct target of this transcription factor ([@pbio-0020355-Bell1]; [@pbio-0020355-Lefebvre2]). Previous studies suggest that SOX9 activity can be modified by protein kinase A (PKA)-dependent protein phosphorylation, or by coexpression of two related proteins, L-SOX5 and SOX6 ([@pbio-0020355-Lefebvre1]; [@pbio-0020355-Huang1]). In addition, close examination of the order of genes induced during chicken digit formation reveals that Sox9 turns on first, followed by Bmpr1b with L-Sox5, and then Sox6 and the cartilage matrix structural components Col2a1 and Agg ([@pbio-0020355-Chimal-Monroy1]). These results, together with the altered pattern of gene expression seen in our Bmpr1a-deficient mice, suggest that BMPR1A signaling may normally act to stimulate SOX9 by post-translational protein modification, or to induce L-Sox5 or Sox6 in cartilage to maintain expression of ECM components. These models are consistent with the ability of BMP2 to both increase PKA activity and induce expression of Sox6 in tissue culture cells ([@pbio-0020355-Lee1]; [@pbio-0020355-Fernandez-Lloris1]). Although we have tried to monitor the expression of L-Sox5 or Sox6 in postnatal articular cartilage, and test the phosphorylation state of SOX9 using previously described reagents ([@pbio-0020355-Lefebvre1]; [@pbio-0020355-Huang1]), we have been unable to obtain specific signal at the late postnatal stages required (unpublished data). Furthermore, null mutations in L-Sox5 or Sox-6 cause lethality at or soon after birth, and no effect on cartilage maintenance has been reported ([@pbio-0020355-Smits1]). However, it seems likely that these or other processes regulated by BMP signaling cooperate with SOX9 to induce target genes in articular cartilage. Mutation of Smad3 or expression of dominant negative transforming growth factor β (TGF-β) type II receptor also disrupts normal articular cartilage maintenance ([@pbio-0020355-Serra1]; [@pbio-0020355-Yang1]). Both manipulations should disrupt TGFβ rather than BMP signaling, and both manipulations cause articular cartilage to hypertrophy and be replaced by bone. In contrast, our analysis of Bmpr1a mutant articular cartilage showed a loss of ECM components, but no signs of hypertrophy or bone replacement. Therefore, TGFβ and BMP signaling are playing distinct but necessary roles to maintain articular cartilage. Although BMPs were originally isolated on the basis of their ability to induce ectopic bone formation, their presence in articular cartilage and strong effect on cartilage formation has stimulated interest in using them to repair or regenerate cartilage defects in adult animals ([@pbio-0020355-Chang1]; [@pbio-0020355-Erlacher1]; [@pbio-0020355-Edwards1]; [@pbio-0020355-Chubinskaya1]). The failure to maintain articular cartilage in the absence of normal BMPR1A function suggests that ligands or small molecule agonists that interact specifically with this receptor subtype may be particularly good candidates for designing new approaches to maintain or heal articular cartilage at postnatal stages. Lack of Bmpr1a function in articular cartilage results in severe fibrillation of the articular surface and loss of joint mobility. The development of severe arthritis symptoms in Bmpr1a-deficient mice raises the possibility that defects in BMP signaling also contribute to human joint disease. Osteoarthritis is known to have a significant genetic component, but it likely involves multiple genetic factors that have been difficult to identify ([@pbio-0020355-Spector1]; [@pbio-0020355-Felson1]; [@pbio-0020355-Hirsch1]). Humans that are heterozygous for loss-of-function mutations in BMPR1A are known to be at risk for juvenile polyposis ([@pbio-0020355-Howe1]; [@pbio-0020355-Zhou1]), but the risk of osteoarthritis for these people has not been reported. However, the control mice used in this study were heterozygous for a null allele of Bmpr1a, and they showed little sign of osteoarthritis even late in life. Several chromosome regions have been previously linked to arthritis phenotypes in humans using either association studies in populations or linkage studies in families. It is interesting to note that several of these chromosome regions contain genes encoding different members of the BMP signaling pathway, including the BMP5 gene on human chromosome 6p12 ([@pbio-0020355-Loughlin1]), the MADH1 gene on human chromosome 4q26--4q31 ([@pbio-0020355-Leppavuori1]; [@pbio-0020355-Kent1]), and the BMPR2 receptor on human chromosome 2q33 ([@pbio-0020355-Wright1]). The complex nature of human osteoarthritis suggests that interactions between multiple genes may be involved in modifying susceptibility to the disease. The inclusion of genetic markers near BMP signaling components may help identify additional osteoarthritis susceptibility loci and facilitate the search for causative mutations. Development and disease processes in synovial joints have been difficult to study genetically, because synovial joints are generated and function at relatively late stages of vertebrate development. The Gdf5-Cre system provides a new method for restricting gene expression or inactivation primarily to articular regions, thus avoiding the pleiotropic functions of many genes in other tissues. Depending on the configuration of the floxed target gene, this system can be used to either activate the expression of a gene primarily in developing joints (ssee [Figure 1](#pbio-0020355-g001){ref-type="fig"}B--[1](#pbio-0020355-g001){ref-type="fig"}D), or to inactivate gene function in articular regions (see [Figure 3](#pbio-0020355-g003){ref-type="fig"}). Additional studies with this system should greatly enhance our knowledge of the development, function, and disease mechanisms of joints, and may bring us closer to better prevention and treatment of joint diseases. Materials and Methods {#s4} ===================== {#s4a} ### Generation of Gdf5-Cre transgenic mice {#s4a1} A mouse 129x1/SvJ BAC library (Invitrogen) was screened to identify a 140-kb BAC from the Gdf5 locus. This BAC was modified using a homologous recombination system in E. coli ([@pbio-0020355-Yang2]) to place nuclear-localized Cre recombinase (from plasmid pML78, gift of Gail Martin) followed by IRES-hPLAP (from plasmid 1726, gift of Oliver Bogler) directly behind the ATG start site of Gdf5. In the process, 583 bp of the first exon of Gdf5 was removed and no functional GDF5 protein is predicted to be produced. The 5′ homology arm was subcloned from a PCR product tailed with XhoI and Bsp120I restriction sites that contains 781 bp of 5′ genomic Gdf5 sequence ending at the ATG translation start site (forward primer 5′-CTGTCTCGAGATGAGGTGGAGGTGAAGACCCC-3′; reverse 5′-GTTTGGGCCCATCCTCTGGCCAGCCGCTG-3′). Cre was subcloned from a 1.1-kb Bsp120I/EcoRI fragment of pML78. IRES hPLAP was subcloned from a 2.1-kb PCR product tailed with EcoRI and SpeI sites that contains the hPLAP translation stop site (forward primer 5′-ATCTCTCGAGGAATTCTCCACCATATTGCCGTCTTTTG-3′; reverse 5′-AGAACTCGAGACTAGTCGGGACACTCAGGGAGTAGTGG-3′). The 3′ homology arm was subcloned from a 0.8-kb PCR product amplified from a 0.9-kb XhoI Gdf5 genomic subclone containing part of the first exon and downstream intron. The forward primer contains the 3′ end of the first exon and is tailed with a SpeI site; the reverse primer is from the T7 promoter of the vector containing the 0.9-kb subclone and flanks the intronic XhoI site (forward primer 5′-CTAAACTAGTCACCAGCTTTATTGACAAAGG-3′; reverse 5′-GATTTCTAGAGTAATACGACTCACTATAGGGC-3′). The targeting construct was built and verified in pBSSK (Stratagene, La Jolla, California, United States), then digested with XhoI and subcloned into pSV1, the vector used for homologous recombination ([@pbio-0020355-Yang2]). Southern blotting, PCR, and DNA sequence analysis confirmed the appropriate targeting construct and BAC modifications were made (unpublished data). Before the modified BAC was injected to produce transgenic animals, a loxP site present in the BAC vector, pBeloBAC11, was removed to prevent the addition of undesired Cre target sites into the genome. To do this, BAC DNA was prepared by CsCl separation, digested with NotI to free the insert from the vector, and size-fractionated over a sucrose gradient. Aliquots of fractions were run on a pulse-field gel and Southern blotted using vector-specific DNA as a probe. Fractions containing unsheared insert and almost no detectable vector DNA were dialyzed in microinjection buffer (10 mM Tris \[pH 7.4\] with 0.15 mM EDTA \[pH 8.0\]) using Centriprep-30 concentrators (Millipore, Billerica, Massachusetts, United States). This purified insert DNA was adjusted to 1 ng/μl and injected into the pronucleus of fertilized eggs from FVB/N mice by the Stanford Transgenic Facility. Transgenic founder mice were identified by PCR using Cre-specific primers 5′-GCCTGCATTACCGGTCGATGCAACGA-3′ and 5′-GTGGCAGATGGCGCGGCAACACCATT-3′, which amplify a 725-bp product, and were assessed for absence of BAC vector using vector-specific primers 5′-CGGAGTCTGATGCGGTTGCGATG-3′ and 5′-AGTGCTGTTCCCTGGTGCTTCCTC-3′, which amplify a 465-bp product. Three lines of Gdf5-Cre mice were established and maintained on the FVB background. Matings with R26R Cre-inducible LACZ reporter mice ([@pbio-0020355-Soriano1]) were used to test for Cre activity. Staining for LACZ and HPLAP on whole embryos or sections of embryos was accomplished following established protocols ([@pbio-0020355-Lobe1]). The red LACZ substrate (see [Figure 1](#pbio-0020355-g001){ref-type="fig"}E) is 6-chloro-3-indoxyl-beta-D-galactopyranoside (Biosynth International, Naperville, Illinois, United States). ### General characterization of Bmpr1a mutant mice {#s4a2} Bmpr1a null and floxed alleles ([@pbio-0020355-Ahn1]; [@pbio-0020355-Mishina3]) were obtained on a mixed 129 and C57BL/6 background and maintained by random breeding. Mice carrying the null and floxed alleles were typically mated to Gdf5-Cre mice as shown in [Figure 3](#pbio-0020355-g003){ref-type="fig"}. The resulting mice are on a mixed 129; C57Bl/6; FVB/N background, with both controls and mutant animals generated as littermates from the same matings. Whole-mount skeletal preparations were made from 34- to 36-d-old mice ([@pbio-0020355-Lufkin1]). Pairs of ears from euthanized 6-mo-old animals were removed, pinned, photographed, projected, and measured from the base of the curve formed between the tragus and antitragus to the farthest point at the edge of the pinnae. Grasping ability in 6-mo-old mice was measured by placing animals on a slender rod and timing how long they could remain suspended on the rod, to a maximum time allowed of 2 min. Data from five consecutive trials for each mouse were averaged. Range of motion assays were conducted on the MT/P1 and P1/P2 joints of the second hindlimb digit from euthanized 18-wk-old animals. Forceps were used to bend the joint to its natural stopping position, and the resulting angle was measured to the nearest 10° under 12.5× magnification using a 360° reticule. Analysis described in this section occurred on animals lacking R26R. Control mice included all nonmutant genotypes generated by Parent 1 being heterozygous for Gdf5-Cre and Bmpr1a^null^ and Parent 2 being heterozygous for Bmpr1a^floxP^ (see [Figure 3](#pbio-0020355-g003){ref-type="fig"}). All statistical analysis used the Student\'s t-test or Welch\'s t-test, and values listed are mean ± standard error of the mean. ### Cell death and proliferation assays {#s4a3} Limbs from mutant and control animals at E13.5 and E14.5 were dissected and frozen in OCT (Sakura Finetek,Torrence, CA, United States). Cryosections of tissue were assayed by TUNEL using the In Situ Cell Death Detection Kit, Fluorescein (Roche, Basel, Switzerland). Following TUNEL, slides were washed in PBS, blocked with PBS + 0.05% Tween-20 + 5% goat serum, washed again, and incubated with a 1:200 dilution of a rabbit anti-phospho-histone-H3 antibody called Mitosis Marker (Upstate Biotechnology, Lake Placid, New York, United States) to identify cells in mitosis. Cy3-labeled anti-rabbit secondary antibody was used to detect the antibody. Cell nuclei were labeled with DAPI, and slides were mounted in Vectamount (Vector Laboratories, Burlingame, California, United States) and visualized at 100× magnification. The area of selected anatomical sites were measured, and the number of TUNEL-labeled nuclear fragments and the number of Cy3-labeled nuclei were counted from three 10-μm sections spanning 50 μm, from three control and three mutant animals. The number of labeled cells in the metacarpal-phalangeal and metatarsal-phalangeal joints was counted in a 290 μm × 365 μm rectangle placed around the center of the joint. The posterior region of the fifth digit was defined by drawing a line from the tip of the digit down 2.15 mm and across to the lateral edge of the tissue. For this analysis, the R26R Cre reporter was not present. ### Histology and histochemistry {#s4a4} Tissue from animals ranging from stages E14.5 to P14 was prepared for analysis by fixing in 4% paraformaldehyde (PFA) in PBS for 45 min to 4 h depending on the stage; washing three times in PBS, once in PBS + 15% sucrose for 1 h, and once in PBS + 30% sucrose for 2 h to overnight depending on the stage; and then freezing in OCT. Tissue from animals aged 7 wk to 9 mo was processed similarly to earlier stages except that it was decalcified in 0.5 M EDTA (pH 7.4) for 4 d prior to incubating in sucrose. All solutions were prechilled and used at 4 °C with agitation, and skin from tissues of P0 or older mice was lacerated or removed prior to processing. Tissue was then cryosectioned at 12 μm and processed. Staining of sections with Safranin O, Fast Green, and Harris\' hematoxylin was carried out using standard histological procedures. Detection of LACZ activity with X-Gal was performed as described ([@pbio-0020355-Lobe1]) and was followed by refixing in 4% PFA, rinsing with deionized water, counterstaining with Nuclear Fast Red (Vector Labs), rinsing with water again, and then mounting in Aquamount (Lerner Labs, Pittsburgh, Pennsylvania, United States). RNA in situ hybridization was performed as described ([@pbio-0020355-Storm1]), with the following modifications: (1) Prior to the acetylation step, sections were incubated with 10--20 μg/ml proteinase K for 30 s to 7 min at room temperature (depending on the developmental stage), followed by refixing in 4% PFA and washing three times in PBS; (2) prehybridization step was skipped, and (3) embryonic tissue sections used a different color development mix ([@pbio-0020355-Thut1]). Probes for the following genes have been published previously: Bmpr1a ([@pbio-0020355-Mishina2]), Col2a1 ([@pbio-0020355-Metsaranta1]), Col10a1 ([@pbio-0020355-Apte1]), Gdf5 ([@pbio-0020355-Storm1]), Osteocalcin ([@pbio-0020355-Celeste1]), and Sox5 and Sox6 ([@pbio-0020355-Lefebvre1]). The following probe templates were gifts: Agg, Dr. Vicki Rosen, Genetics Institute; Bmp2 and Bmp4, Arend Sidow, Stanford University; Col1a1, Bjorn Olsen, Harvard Medical School; Bmpr1b, Col3a1, and Mat4 probes were made from ESTs with IMAGE clone numbers 5056341, 478480, and 406027, respectively (Invitrogen, Carlsbad, California, United States). Sections for immunohistochemistry were fixed in 4% PFA, then digested with 942--2,000 U/ml type IV-S bovine hyaluronindase (Sigma, St. Louis, Missouri, United States) in PBS (pH 5) at 37 °C for 30 min to 2 h depending on the stage. Slides were then washed in PBS, treated with 0.3% hydrogen peroxide in 100% methanol for 30 min, washed, blocked with PBS + 0.05% Tween20 + 5% goat or fetal bovine serum, washed again, and incubated with primary antibodies in PBS + 0.05% Tween 20 + 1% goat or fetal bovine serum overnight at 4 °C. Biotin-labeled secondary antibodies (Vector Labs) were tagged with HRP using the Vectastain Elite ABC kit (Vector Labs) followed by detection with DAB (Vector Labs). Primary antibodies and dilutions used were: goat anti-mouse MMP13, 1:100 (Chemicon International, Temecula, California, United States); rabbit anti-human SOX9, 1:500 ([@pbio-0020355-Morais1]); rabbit anti-phosphorylated-SOX9 (SOX9.P), 1:10--1:250 ([@pbio-0020355-Huang1]). Supporting Information {#s5} ====================== Accession Numbers {#s5a1} ----------------- GenBank (<http://www.ncbi.nih.gov/Genbank/>) accession numbers for the genes discussed in this paper are Gdf5 (AC084323) and Bmpr1a (NM\_009758). We thank Gail Martin for the Cre construct (plasmid pML78) and Oliver Bogler for the IRES-hPLAP construct (plasmid 1726); Bjorn Olsen and Benoit de Crombrugghe for antibodies; the following individuals for in situ probe templates: Sophie Candille, Arend Sidow (Bmp2 and Bmp4), Vicki Rosen (Agg), and Bjorn Olsen (Col1a1); Véronique Lefebvre for Sox5 and Sox6 probe templates and useful discussions; Michelle Johnson for help with phenotypic assays on mice; Dr. Corrine Davis for help in evaluating synovial sections; Rebecca Rountree for Adobe Photoshop and Illustrator tips; and members of the Kingsley lab for helpful comments on the manuscript. This work was supported by an NIH predoctoral training grant (RR), a postdoctoral fellowship from the Arthritis Foundation (MS), and grants from the National Institutes of Health (DK). Dr. Kingsley is an associate investigator of the Howard Hughes Medical Institute. Conflicts of interest. The authors have declared that no conflicts of interest exist. Author contributions. RBR, MS, and DMK conceived and designed the experiments. RBR and MEM performed the experiments. RBR, HC, and DMK analyzed the data. MS, HC, VH, and YM contributed reagents/materials/analysis tools. RBR and DMK wrote the paper. Academic Editor: Lee Niswander, University of Colorado Health Sciences Center ¤ Current address: ARTEMIS Pharmaceuticals, an Exelixis Company, Köln, Germany Citation: Rountree RB, Schoor M, Chen H, Marks ME, Harley V, et al. (2004) BMP receptor signaling is required for postnatal maintenance of articular cartilage. PLoS Biol 2(11): e355. BAC : bacterial artificial chromosome Bmpr1a : bone morphogenetic protein receptor 1a E\[number\] : embyonic day \[number\] ECM : extracellular matrix GAC : transgenic line carrying Gdf5-alkaline phosphatase-Cre construct Gdf5 : growth differentiation factor 5 hPLAP : human placental alkaline phosphatase IRES : internal ribosome entry site PFA : paraformaldehyde R26R : lacZ ROSA26 Cre reporter strain TGF-β : transforming growth factor β TUNEL : terminal deoxynucleotidyl transferase--mediated deoxyuridine triphosphate nick end labeling	PubMed Central	2024-06-05T03:55:48.558648	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523229/", "journal": "PLoS Biol. 2004 Nov 19; 2(11):e355", "authors": [ { "first": "Ryan B", "last": "Rountree" }, { "first": "Michael", "last": "Schoor" }, { "first": "Hao", "last": "Chen" }, { "first": "Melissa E", "last": "Marks" }, { "first": "Vincent", "last": "Harley" }, { "first": "Yuji", "last": "Mishina" }, { "first": "David M", "last": "Kingsley" } ] }
PMC523230	Introduction {#s1} ============ The Assembly of Bipolar Thick Filaments Is Regulated During Cell Division {#s1a} ------------------------------------------------------------------------- Myosin-II (hereafter referred to as myosin) is a hexameric protein composed of two heavy chains, two regulatory light chains, and two essential light chains ([Figure 1](#pbio-0020356-g001){ref-type="fig"}A). The heavy chain consists of an N-terminal globular head that contains ATP and actin-binding sites, an α-helical neck that contains the light-chain binding sites, and finally a long α-helix that dimerizes with the α-helix of the other heavy chain to form the coiled-coil tail. ::: {#pbio-0020356-g001 .fig} Figure 1 ::: {.caption} ###### Domains and Charge Distribution within the Myosin Tail \(A) The myosin head (1--818) and light chains are shown at the N terminus. In the coiled-coil tail, Ala 1 is red (1,348--1,530), the AD is blue (1,531--1,824), Ala 2 is purple (1,825--1,966), the C-terminal domain is green (1,967--2,116), and the remainder is black (819--1,347). Phospho-threonines (at positions 1,823, 1,833, and 2,029) are indicated by the letter T. (B and C) Plots of the average charge of each tail domain color coded as in (A). The y-axis is average charge; the x-axis is tail position. Aspartic acid and glutamic acid are assigned --1, lysine and arginine are assigned +1, and all other a.a. are assigned 0. The average charge in (B) was determined with a window size of 14 a.a., and the average charge in (C) was determined with a window size of 28 a.a.. Arrows highlight the 28 a.a. charge repeat in (B) and the 196 a.a. charge repeat in (C). \(D) "Headless" AD-Cterm (3xThr) (1,531--2,116). ::: ![](pbio.0020356.g001) ::: A bipolar thick filament (BTF) is a highly organized structure composed of individual myosin molecules. In vitro, myosin from the cellular slime mold Dictyostelium discoideum assembles without the aid of a cofactor, demonstrating that the myosin molecule itself contains all the information needed to form a highly organized structure of a defined size ([@pbio-0020356-Clarke1]). While proteins that co-assemble with muscle myosin have been discovered in some cell types, no such proteins have been identified in Dictyostelium ([@pbio-0020356-Barral1]). In vivo, BTFs are important in a variety of cellular processes, including cell motility, chemotaxis, and development. In cytokinesis, BTFs provide the mechanical force to constrict an actin ring positioned at the midzone of the dividing cell ([@pbio-0020356-De1]). In Dictyostelium, phosphorylation of myosin inhibits filament assembly ([@pbio-0020356-Kuczmarski1]). Myosin is recruited to the cleavage furrow in the form of a BTF ([@pbio-0020356-Sabry1]). After the BTF provides contractile force, a myosin heavy chain kinase is recruited that then drives the disassembly of the BTF ([@pbio-0020356-Liang2]). Phosphorylation of only three threonines in the 2,116-amino acid (a.a.) heavy chain of myosin is sufficient to inhibit BTF formation ([@pbio-0020356-Vaillancourt1]; [@pbio-0020356-Luck-Vielmetter1]). These threonines have been mutated to aspartic acid (3xAsp myosin) to create a mimic of a fully phosphorylated state ([@pbio-0020356-Egelhoff1]). 3xAsp myosin is assembly-incompetent in vitro, and the phenotype of Dictyostelium cells expressing only 3xAsp myosin is similar to the myosin-null mutant, consistent with previous work demonstrating that BTF assembly is required for myosin function in vivo ([@pbio-0020356-De3]; [@pbio-0020356-Knecht1]; [@pbio-0020356-Manstein1]; [@pbio-0020356-Egelhoff1]). Mutation of the phosphorylation sites to alanines (3xAla myosin), in contrast, does not alter the filament formation properties of myosin in vitro. In vivo, 3xAla myosin is always assembled and therefore is a good mimic of an unphosphorylated state ([@pbio-0020356-Egelhoff1]; [@pbio-0020356-Yumura1]). The doubling time of 3xAla cells is 17 h, whereas in wild-type cells it is 13 h, suggesting that thick filament disassembly is required for efficient cytokinesis ([@pbio-0020356-Egelhoff1]). The Role of Charge Repeats in the Myosin Tail {#s1b} --------------------------------------------- Myosin tails have a striking pattern of charged a.a., with an average positive charge over 14 a.a. followed by an average negative charge over 14 a.a. to form a 28-a.a. charge repeat throughout the tail ([@pbio-0020356-McLachlan2]; [@pbio-0020356-McLachlan1]). A second charge repeat found only in the C-terminal 68 kDa of the Dictyostelium myosin tail occurs every 196 a.a. ([@pbio-0020356-Warrick1]; [@pbio-0020356-Shoffner1]). These repeating patterns can be visualized by considering the tail as a one-dimensional rod and calculating the average charge over a window of a.a. as a function of position in the tail ([@pbio-0020356-Shoffner1]). A 14-a.a. window makes the 28-a.a. repeat apparent ([Figure 1](#pbio-0020356-g001){ref-type="fig"}B), and a 28-a.a. window averages out the 28-a.a. pattern, making the 196 a.a. pattern more apparent ([Figure 1](#pbio-0020356-g001){ref-type="fig"}C). The Myosin Tail Contains Functional Domains {#s1c} ------------------------------------------- The C-terminal 68 kDa of the tail contains the 34-kDa assembly domain (AD; a.a. 1,531--1,824; [Figure 1](#pbio-0020356-g001){ref-type="fig"}, blue coiled-coil). The AD has salt-dependent solubility properties like full-length, wild-type myosin in vitro ([@pbio-0020356-OaHalloran1]; [@pbio-0020356-Shoffner1]). However, unlike full-length myosin, which assembles into BTFs, the AD assembles into paracrystals having undefined size and number of elements. The AD also appears to be the minimal assembling portion of the Dictyostelium myosin tail. C-terminal truncation of full-length myosin to a.a. 1,819 yields an assembly-competent myosin, whereas deletion of an additional 35 a.a. abolishes assembly ([@pbio-0020356-Lee1]). The Role of a Globular Head in Assembly {#s1d} --------------------------------------- Replacement of the catalytic domain and essential light chain binding domain of full-length Dictyostelium myosin with green fluorescent protein (GFP) produces a chimera that is apparently indistinguishable from full-length myosin in its ability to form BTFs ([@pbio-0020356-Zang1]). Therefore, the myosin head is not required for BTF formation. However, the contribution of a globular head to the assembly process has not been closely examined. Models of Regulation {#s1e} -------------------- Several models have been proposed for the mechanism of myosin regulation. In this discussion, the hexameric myosin molecule will be referred to as a monomer ([Figure 1](#pbio-0020356-g001){ref-type="fig"}A). A monomer-sequestering model was hypothesized based on the characterization of 3xAsp myosin in vitro by rotary shadowing electron microscopy ([@pbio-0020356-Pasternak1]; [@pbio-0020356-Liang1]) and in vivo by identification of myosin tail mutations that suppressed the 3xAsp phenotype ([@pbio-0020356-Liang1]). In this model, the coiled-coil tail bends and folds back on itself, sequestering the AD from other myosin molecules. It was further hypothesized that the bent conformation is stabilized by an intramolecular interaction between two regions of the tail rich in alanines in core positions of the heptad repeat. These coiled-coil regions, termed Ala 1 (a.a. 1,348--1,530) and Ala 2 (a.a. 1,825--1,966), are shown in red and purple respectively in [Figure 1](#pbio-0020356-g001){ref-type="fig"}. A bias for alanines in core heptad positions is characteristic of antiparallel tetrameric coiled-coils ([@pbio-0020356-Munson1]), prompting the speculation that Ala 1 and Ala 2 form an antiparallel tetrameric coiled-coil in a phosphorylation-dependent manner. An alternative monomer-sequestering mechanism might be destabilization of the coiled-coil. Threonine 1,823 is located at the C terminus of the AD, and is predicted to be in the D position of the heptad repeat. The D position is in the core of the coiled-coil where the two strands interact closely. It is possible that phosphorylation of threonine 1,823 would cause a local disruption of the coiled-coil and perturb self-assembly by introducing negative charge into the hydrophobic core ([@pbio-0020356-Luck-Vielmetter1]; [@pbio-0020356-Liang1]; [@pbio-0020356-Nock1]). Consistent with this hypothesis, [@pbio-0020356-Nock1] showed that in full-length myosin, position 1,823 makes the largest contribution to the 3xAsp phenotype. However, the susceptibility of the myosin tail to proteolysis by chymotrypsin does not change with phosphorylation, arguing that large structural changes due to phosphorylation are unlikely ([@pbio-0020356-Cote1]). Another model postulates that regulation occurs at the level of assembly intermediates. The assembly pathway of full-length Dictyostelium myosin consists of a nucleation phase followed by an elongation phase ([@pbio-0020356-Mahajan1]). In the nucleation phase, a parallel dimer forms in which two myosin monomers self-associate in a parallel orientation with a 14-nm stagger. When two myosin monomers are offset by 14 nm, both the 28-a.a. and 196-a.a. charge repeats are aligned to maximize charge complementation between the two tails ([@pbio-0020356-De2]). Threonine 1,823 is positioned within a cluster of positively charged a.a. that forms part of the 196 a.a. repeat ([@pbio-0020356-Luck-Vielmetter1]; [@pbio-0020356-Nock1]). The introduction of negative charge by phosphorylation of threonine 1,823 could disrupt critical charge-charge interactions required for assembly. Localization Models {#s1f} ------------------- Determining the mechanism of regulated assembly will help elucidate the largely uncharacterized mechanism of myosin recruitment to the cleavage furrow. Reports that 3xAsp myosin does not localize, while 3xThr (i.e., wild-type) and 3xAla can localize, have demonstrated the necessity of BTF assembly for proper localization of myosin ([@pbio-0020356-Sabry1]). However, the motor activity is not required, as the chimeric GFP-regulatory light-chain tail protein is capable of translocation to the cleavage furrow in dividing Dictyostelium cells ([@pbio-0020356-Yumura2]; [@pbio-0020356-Zang1]). Chimeric constructs consisting of the Dictyostelium motor domain and the Acanthamoeba, chicken smooth muscle, or skeletal muscle myosin tails translocate to the cleavage furrow of Dictyostelium cells despite almost no sequence homology between these tails and the Dictyostelium myosin tail ([@pbio-0020356-Shu2]; [@pbio-0020356-Shu3]). This has led to the hypothesis that no specific a.a. sequence is required, but that assembly is both necessary and sufficient for cleavage furrow localization ([@pbio-0020356-Shu2]; [@pbio-0020356-Shu3]; [@pbio-0020356-Shu1]). Reconstitution of Regulated BTF Assembly {#s1g} ---------------------------------------- Characterization of Dictyostelium BTFs has identified functional domains and repeating patterns in the coiled-coil tail. How do these properties come together to form a BTF that can be regulated by phosphorylation of only three a.a.? How do these properties contribute to regulated assembly? How do these properties enable a myosin BTF to localize to the cleavage furrow during cytokinesis? We have approached these questions by reconstituting regulated assembly of BTFs and defining the minimal part of the molecule required for regulated BTF assembly in vitro and in vivo. Results {#s2} ======= The AD-Cterm (3xThr) and (3xAsp) Tail Fragments Reconstitute Regulated Assembly {#s2a} ------------------------------------------------------------------------------- We wished to test whether both the Ala 1 and the Ala 2 domains of myosin are necessary to regulate BTF assembly ([@pbio-0020356-Liang1]) or whether the AD-through-C-terminal portion of the tail is sufficient for formation of regulated BTFs. We constructed a tail fragment that starts at the AD and extends to the end of the C terminus (AD-Cterm) ([Figure 1](#pbio-0020356-g001){ref-type="fig"}D), and therefore contains all three threonine phosphorylation sites but not the Ala1 region. We created both "wild-type" (AD-Cterm \[3xThr\]) and "3xAsp" (AD-Cterm \[3xAsp\]) versions; the latter construct had aspartic acids in place of the three threonine phosphorylation sites to mimic a constitutively phosphorylated state. We then compared the salt-dependent solubility of these tail fragments to full-length phosphorylated and unphosphorylated myosin to see whether AD-Cterm recapitulates regulated assembly. Thick filaments and paracrystals efficiently sediment upon centrifugation, whereas unassembled molecules remain soluble. Wild-type full-length myosin efficiently assembles in buffers of intermediate ionic strength (25--100 mM NaCl), while phosphorylated full-length myosin is unassembled ([Figure 2](#pbio-0020356-g002){ref-type="fig"}A) ([@pbio-0020356-Kuczmarski1]; [@pbio-0020356-Cote1]). Therefore, regulated assembly is biochemically defined as salt-dependent insolubility of unphosphorylated myosin and solubility of phosphorylated myosin. AD-Cterm (3xThr) and AD-Cterm (3xAsp) possess the same in vitro assembly properties as full-length unphosphorylated and full-length phosphorylated myosin, respectively, suggesting that this part of the tail contains all the information needed to regulate assembly in vitro ([Figure 2](#pbio-0020356-g002){ref-type="fig"}A). ::: {#pbio-0020356-g002 .fig} Figure 2 ::: {.caption} ###### Characterization of "Headless" AD-Cterm Tail Fragments \(A) Analysis of assembly by sedimentation. Fraction of soluble protein as a function of NaCl concentration is plotted for the constructs depicted adjacent to the graph. The solubility of "headless" AD-Cterm (3xThr) and AD-Cterm (3xAsp) are compared to the solubility of unphosphorylated and phosphorylated full-length myosin having the globular motor domain (full-length myosin data from [@pbio-0020356-Cote1]). \(B) Sedimentation equilibrium analysis of 52 μM "headless" AD-Cterm (3xThr) and AD-Cterm (3xAsp) in buffer containing 500 mM NaCl. The top graphs show the concentration distribution, fit, and residuals for AD-Cterm (3xThr), while the bottom graphs show the same data for AD-Cterm (3xAsp). The molecular weight obtained from the fit was 130 kDa for AD-Cterm (3xThr) and 120 kDa for AD-Cterm (3xAsp). \(C) Thermal melts of "headless" AD-Cterm (3xThr) and AD-Cterm (3xAsp) in buffer containing 500 mM NaCl are shown as fraction of protein denatured as a function of temperature. The open circles are data for 50 μM "headless" AD-Cterm (3xThr), and the open squares are data for 50 μM "headless" AD-Cterm (3xAsp). ::: ![](pbio.0020356.g002) ::: The Global Stabilities of AD-Cterm (3xThr) and AD-Cterm (3xAsp) Are Identical {#s2b} ----------------------------------------------------------------------------- We next tested whether AD-Cterm (3xThr) and AD-Cterm (3xAsp) tail fragments form coiled-coils, similar to full-length myosin. The oligomerization state of the tail fragments was determined by sedimentation equilibrium analysis, and the secondary structure was assessed by circular dichroism (CD). Neither tail fragment is expected to assemble in the buffers used for these studies, as they contain 500 mM NaCl. Sedimentation runs were carried out as a function of protein concentration, and equilibrium traces were obtained after 20 h of centrifugation. The profiles were fit to an equation describing the sedimentation behavior of a single, non-associating, ideal species ([Figure 2](#pbio-0020356-g002){ref-type="fig"}B). The distribution of residuals around 0 suggests that the concentration distribution is well described by the model. The predicted molecular weight of the coiled-coil is 137 kDa, and the experimentally determined molecular weight ranged from 130 kDa to 160 kDa for the AD-Cterm (3xThr) tail fragment and 120 kDa to 150 kDa for the AD-Cterm (3xAsp) tail fragment. The range of molecular weights obtained for both tail fragments was due to an observed decrease in molecular weight as protein concentration was increased. This trend is a hallmark of non-ideality and has been observed for elongated, rod-like molecules such as DNA and skeletal muscle myosin ([@pbio-0020356-Tanford1]). Far-UV CD spectra at 4 °C show that both tail fragments are α-helical with similar α-helical content as assessed by comparing mean residue ellipticity at 222 nm (θ~222~) (unpublished data). Therefore, both tail fragments behave as two-stranded coiled-coils in these assays. The sedimentation equilibrium analysis demonstrates that a direct comparison of coiled-coil stability can be made by thermal denaturation at 500 mM NaCl because neither tail fragment self-assembles to form larger species in this condition. Thermal melts are reversible and at equilibrium and far-UV CD spectra show both tail fragments are α-helical at the starting temperature of the melt (4 °C) and are random coil at the ending temperature (60 °C) (unpublished data). The melting temperature for the AD-Cterm (3xThr) and AD-Cterm (3xAsp) tail fragments at both protein concentrations are identical (29 °C at 50 μM and 28 °C at 2.5 μM; [Figure 2](#pbio-0020356-g002){ref-type="fig"}C). In summary, the AD-Cterm tail fragments behave as two-stranded coiled coils with similar thermal denaturation properties in high salt ([Figure 2](#pbio-0020356-g002){ref-type="fig"}B and [2](#pbio-0020356-g002){ref-type="fig"}C). Therefore, the difference in assembly observed between the wild-type and 3xAsp tail fragments cannot be attributed to a failure of AD-Cterm (3xAsp) to fold into an α-helical two-stranded coiled-coil. AD-Cterm Tail Fragments That Have a GFP "Head" Form BTFs {#s2c} -------------------------------------------------------- The AD-Cterm data show that tail fragment solubility accurately predicts the bulk solubility properties of full-length myosin. However, they do not form true BTFs. When AD-Cterm (3xThr) is assembled in buffer containing 50 mM NaCl with 10 mM MgCl~2~ and imaged using negative-stain (EM), it forms paracrystals with a 14-nm (98-a.a.) periodicity, corresponding to the periodicity of charge in the AD-through-C-terminal region of the myosin tail ([@pbio-0020356-De4]; [@pbio-0020356-OaHalloran1]) ([Figure 3](#pbio-0020356-g003){ref-type="fig"}A). ::: {#pbio-0020356-g003 .fig} Figure 3 ::: {.caption} ###### Analysis of "Headless" AD-Cterm (3xThr) and GFP-AD-Cterm (3xThr) Assembly by EM (A and B) The scale bars are 100 nm, and all panels are on the same scale. (A) The "headless" AD-Cterm (3xThr) tail fragment assembled for 2 h. (B) Three images of GFP-AD-Cterm (3xThr) assembled for 2--5 min. \(C) Analysis of GFP AD-Cterm (3xThr) (open circles) and GFP-AD-Cterm (3xAsp) (open squares) assembly by sedimentation. ::: ![](pbio.0020356.g003) ::: To test whether the presence of a globular domain might induce AD-Cterm to form regulated, bipolar structures analogous to the thick filaments formed by full-length Dictyostelium myosin, we attached a GFP molecule to the N terminus of AD-Cterm (GFP-AD-Cterm). Both GFP-AD-Cterm (3xThr) and GFP-AD-Cterm (3xAsp) versions of this protein were made to mimic the unphosphorylated and phosphorylated states of myosin, respectively. We assembled purified GFP-AD-Cterm (3xThr) in buffer containing 50 mM NaCl with 10 mM MgCl~2~ and imaged them using EM. Bipolar structures first formed on a time scale of 2--5 min ([Figure 3](#pbio-0020356-g003){ref-type="fig"}B), then bound together and reorganized to form larger, but less well-defined structures on a time scale of more than 10 min. This result is consistent with previous data on full-length Dictyostelium myosin indicating that when it is prepared by dialysis rather than by rapid dilution as used here (see [Materials and Methods](#s4){ref-type="sec"}), it forms elongated structures ([@pbio-0020356-Stewart1]). The bipolar structures have striations at their ends spaced 14.4 ± 2.9 nm (n = 25) apart, corresponding well to the 14.3-nm offset of full-length myosin heads in a BTF ([@pbio-0020356-Stewart1]). GFP-AD-Cterm (3xThr) BTFs Are Structurally Homologous to Full-Length Myosin BTFs {#s2d} -------------------------------------------------------------------------------- We compared the dimensions of GFP-AD-Cterm (3xThr) and full-length myosin BTFs ([@pbio-0020356-Clarke1]; [@pbio-0020356-Stewart1]). The width of GFP-AD-Cterm (3xThr) BTFs is 27 ± 6 nm (n = 63) at the center of the bare zone (area where heads are absent), close to the width of full-length myosin BTFs (33 ± 1 nm). For both GFP-AD-Cterm and full-length myosin, the length of the bare zone is approximately equal to the length of the coiled-coil. This length is 130--190 nm for full-length myosin and 85 ± 11 nm (n = 63) for GFP-AD-Cterm. This is an indication that GFP-AD-Cterm probably assembles in a manner similar to full-length myosin, with differences in BTF dimensions reflecting differences in myosin tail length. Notably, the solubility properties of GFP-AD-Cterm (3xThr) and GFP-AD-Cterm (3xAsp) ([Figure 3](#pbio-0020356-g003){ref-type="fig"}C) are similar to the headless AD-Cterm (3xThr) and AD-Cterm (3xAsp) tail fragments, respectively, as well as to unphosphorylated and phosphorylated full-length myosin-II, respectively (see [Figure 2](#pbio-0020356-g002){ref-type="fig"}A). The AD-Cterm Tail Fragment Is Sufficient for Regulated Localization of Myosin {#s2e} ----------------------------------------------------------------------------- To test the ability of GFP-tail fragments to localize to the cleavage furrow of Dictyostelium in vivo, we expressed our constructs in myosin-null Dictyostelium cells. However, the GFP-tail fragments were vastly overexpressed in vivo (unpublished data). Introducing the 31-a.a. regulatory light chain (RLC) binding site between the GFP and tail fragment sequences ([@pbio-0020356-Zang1]) produced expression levels similar to wild-type myosin (unpublished data). We expressed 3xThr, 3xAla, and 3xAsp versions of GFP-RLC-AD-Cterm in myosin heavy chain-null Dictyostelium cells. To ensure that the RLC did not interfere with filament formation, we assayed for assembly of the GFP-RLC-tail fusions in Dictyostelium cell extracts. As expected, the 3xThr and 3xAla GFP-RLC-AD-Cterm proteins were assembly-competent in Dictyostelium extracts, while GFP-RLC-AD-Cterm (3xAsp) was not (unpublished data), demonstrating that the RLC does not interfere with regulated filament formation. We used live-cell fluorescence microscopy to study the localization of the GFP-RLC-tail fragments in Dictyostelium cells ([Figure 4](#pbio-0020356-g004){ref-type="fig"}). In 9 of 9 dividing cells, GFP-RLC-AD-Cterm (3xAla) localized to the cleavage furrow and remained at the site of cleavage furrow formation in the resulting daughter cells (termed "back end") after cytokinesis. In contrast, GFP-RLC-AD-Cterm (3xAsp) did not go to the cleavage furrow in any of 11 observed dividing cells. GFP-RLC-AD-Cterm (3xThr) localized to the furrow in 6 of 12 dividing cells, and in these six cells localization was apparent only in the very late stage of furrow formation and at the back end of the resulting daughter cells. In three cells, GFP-RLC-AD-Cterm (3xThr) localization was clearly visible only at the back end of the resulting daughter cells, and in the remaining three cells no localization was clear. Thus, the AD-Cterm tail fragment is sufficient for localization, but does not localize as efficiently as full-length GFP-myosin (3xThr). The increased localization of GFP-RLC-AD-Cterm (3xAla) is consistent with the overassembly and increased localization reported for full-length GFP-myosin (3xAla) ([@pbio-0020356-Sabry1]; [@pbio-0020356-Robinson1]). ::: {#pbio-0020356-g004 .fig} Figure 4 ::: {.caption} ###### Localization of GFP-RLC-Tail Fragment Constructs in Live Dividing Dictyostelium Cells The localization of several GFP-RLC-myosin tail fragments during and just after cytokinesis in live Dictyostelium cells are shown. GFP-myosin (row 1) is clearly localized to the early and late cleavage furrow of the dividing cell and to the back end of the resulting daughter cells. By contrast, GFP-RLC-AD-Cterm (3xThr) (row 2) is localized correctly only at the late stages of cytokinesis and in the back end of one daughter cell. GFP-RLC-AD-Cterm (3xAla) (row 3) is localized to the furrow as well as to the back end of a daughter cell, while GFP-RLC-AD-Cterm (3xAsp) (row 4) shows diffuse localization throughout cytokinesis. The scale bar is 10 μm and the time is indicated in min:sec. ::: ![](pbio.0020356.g004) ::: Objects of Comparable Length to BTFs Are Not Enriched at the Cleavage Furrow During Cytokinesis {#s2f} ----------------------------------------------------------------------------------------------- To examine the specificity of myosin BTF localization, we examined whether any object of comparable length to a BTF is enriched at the cleavage furrow during cytokinesis ([@pbio-0020356-Uyeda1]). We scrape-loaded 0.5 μm diameter fluorescent beads into Dictyostelium cells expressing GFP fused to the N terminus of full-length wild-type myosin. These are round beads, while Dictyostelium BTFs are rod-shaped, but the length of each structure is comparable (approximately 0.5 μm). [Figure 5](#pbio-0020356-g005){ref-type="fig"}A shows a time course of a representative Dictyostelium cell during cytokinesis. In these cells, GFP-myosin is recruited to the cell equator early during cytokinesis and remains there until the cell divides. Thereafter, myosin remains at the back end of the daughter cell as they move away from one another. In contrast, the labeled beads of similar size show no directed motion toward the furrow ([Figure 5](#pbio-0020356-g005){ref-type="fig"}B). ::: {#pbio-0020356-g005 .fig} Figure 5 ::: {.caption} ###### Localization of 0.5-μm Beads in Live Dividing Dictyostelium Cells \(A) Time course of a representative GMO8B Dictyostelium cell (defined in [Materials and Methods](#s4){ref-type="sec"}; contains GFP myosin) during cytokinesis. GFP-myosin fluorescence is shown in green and the beads are shown in red. While GFP-myosin accumulates in the furrow, the beads show no directed motion. The scale bar is 10 μm and time is indicated in min:sec. \(B) Plot of the location of each bead in each of six dividing cells in the six frames imaged during cytokinesis. The axes were defined in each frame to bisect the center of the cell both horizontally and vertically such that the center of the furrow is the origin of the axes. The position of each bead was then plotted relative to these axes. Different beads are represented by different colors. For three of the beads, the trajectory of the bead is shown with arrows. The average location of a cell is outlined on the plot in transparent green. ::: ![](pbio.0020356.g005) ::: Are All the Domains within the AD-Cterm Tail Fragment Required for Regulated Assembly {#s2g} ------------------------------------------------------------------------------------- The data detailed in [Figures 2](#pbio-0020356-g002){ref-type="fig"}--[4](#pbio-0020356-g004){ref-type="fig"} show that the tail fragments examined in this study are a good model for a myosin BTF. To examine the roles of the various domains within this tail fragment, we have constructed several shorter fragments of the myosin tail ([Figure 6](#pbio-0020356-g006){ref-type="fig"}) and analyzed their regulated assembly properties. ::: {#pbio-0020356-g006 .fig} Figure 6 ::: {.caption} ###### Tail Fragments Used for Truncation Analysis For tail fragments that include more than one domain, the name is determined by the first and last domain in the tail fragment. Phosphorylation sites are indicated in parentheses. 1xThr indicates that the fragment contains the threonine at a.a. 1,823; 2xThr indicates that the fragment contains the threonines at a.a. 1,823 and 1,833; and 3xThr indicates that the fragment contains the threonines at a.a. 1,823, 1,833, and 2,029. The same scheme is used to describe aspartic acid-containing constructs in the paper, except threonine is substituted with aspartic acid. ::: ![](pbio.0020356.g006) ::: The AD Does Not Reconstitute Regulated Assembly {#s2h} ----------------------------------------------- Given the large contribution that a.a. 1,823, the penultimate a.a. in the AD, makes to the 3xAsp phenotype ([@pbio-0020356-Nock1]), we tested whether the AD could be regulated by this a.a. alone. The salt-dependent solubilities of the AD (1xThr) and a mutant in which threonine 1,823 had been changed to aspartic acid (AD \[1xAsp\]) were nearly identical in the sedimentation assay ([Figure 7](#pbio-0020356-g007){ref-type="fig"}A), which suggests that the AD can assemble but is not sufficient for regulated assembly. ::: {#pbio-0020356-g007 .fig} Figure 7 ::: {.caption} ###### Analysis of Assembly by Sedimentation The solubility of the various constructs used for the truncation analysis is compared. \(A) Comparison of "headless" AD (1xThr) and "headless" AD (1xAsp). \(B) Comparison of "headless" AD (1xThr), GFP-AD (1xThr), and GFP-AD-Cterm (3xThr). The GFP is located at the N-terminus in both GFP-containing constructs. \(C) Comparison of "headless" extended AD (2xThr) and "headless" extended AD (2xAsp). \(D) Comparison of "headless" AD-Ala2 (2xThr) and "headless" AD-Ala2 (2xAsp). \(E) Comparison of "headless" AD-Cterm (3xThr), "headless" Ala1-Ala2 (2xThr), and "headless" Ala1-Cterm (3xThr). ::: ![](pbio.0020356.g007) ::: We attached a GFP "head" to the AD (1xThr) and, interestingly, this GFP-AD (1xThr) tail fragment had bulk sedimentation properties very similar to those of headless AD (1xThr), suggesting that the presence of a globular domain does not affect in vitro solubility, similar to AD-Cterm and GFP-AD-Cterm tail fragments ([Figure 7](#pbio-0020356-g007){ref-type="fig"}B). The AD (1xThr) tail fragments are more salt-sensitive than AD-Cterm (3xThr) tail fragments, possibly because they have a shorter coiled-coil tail ([Figure 7](#pbio-0020356-g007){ref-type="fig"}A and [7](#pbio-0020356-g007){ref-type="fig"}B). EM showed that GFP-AD (1xThr) formed structures of fixed size that were not true BTFs because they did not contain bare zones (unpublished data). This demonstrates that a globular head fixes the size of assembling myosin structures, while C-terminal sequence elements in the myosin tail may be required for proper formation of the bare zone. Regulated Assembly Cannot Be Reconstituted with Tail Fragments Smaller Than AD-Cterm {#s2i} ------------------------------------------------------------------------------------ To determine whether the entire 34-kDa portion of the tail C-terminal to the AD is required for regulated assembly, we generated a series of tail fragments that all start at the AD and are truncated at different positions within the C-terminal 34 kDa of the tail (see [Figure 6](#pbio-0020356-g006){ref-type="fig"} for constructs). The first such fragments that we examined started at the AD and included threonine 1,833, the second phosphorylation site (extended AD \[2xThr\]) and extended AD \[2xAsp\]). Unlike AD (1xThr) and AD (1xAsp), extended AD (2xThr) and extended AD (2xAsp) show a four-fold difference in solubility ([Figure 7](#pbio-0020356-g007){ref-type="fig"}C). This is significantly less than the eight-fold difference in solubility of full-length unphosphorylated and phosphorylated myosin, respectively, and AD-Cterm (3xThr) and AD-Cterm (3xAsp), respectively. To test whether adding on a larger portion of the tail results in a greater degree of regulated assembly, we constructed tail fragments that start at the AD and end at a.a. 1,966, the end of Ala 2 (AD-Ala 2). As with other tail fragments, the aspartic acid variant of this construct is more soluble than its wild-type counterpart ([Figure 7](#pbio-0020356-g007){ref-type="fig"}D). Surprisingly, AD-Ala 2 (2xThr) exhibits inhibited assembly relative to shorter constructs ([Figure 7](#pbio-0020356-g007){ref-type="fig"}D). This inhibition was not merely a consequence of where we truncated this tail fragment. A tail fragment was constructed that started at the AD and ended at a.a. 2,015 (AD-2015 \[2xThr\]). This construct contains an isoleucine in the C-terminal most heptad core position, making it less likely that the two strands of the coiled-coil will fray. Like AD-Ala 2 (2xThr), AD-2015 (2xThr) exhibits inhibited self-assembly (unpublished data). Furthermore, a far-UV CD spectrum of AD-Ala2 (2xThr) showed that it is α-helical (unpublished data), indicating that AD-Ala2 (2xThr) is structurally intact. All tail fragments truncated in the C-terminal 34 kDa of the tail exhibit altered assembly properties, while AD-Cterm (3xThr) and AD-Cterm (3xAsp) possess the same in vitro assembly properties as full-length, unphosphorylated, and phosphorylated myosin, respectively (see [Figure 7](#pbio-0020356-g007){ref-type="fig"}A) ([@pbio-0020356-Cote1]). These data suggest that the entire AD-Cterm tail fragment is required to reconstitute regulated assembly of BTFs. The Ala 1 Domain Stabilizes Assembly {#s2j} ------------------------------------ When the entire C-terminal domain is included in the AD-Cterm tail fragment, the inhibitory effect of Ala 2 is overcome. To test whether this stabilizing effect is specific to the C-terminal domain, we determined if Ala 1 could also stabilize assembly. A tail fragment was constructed that starts at Ala 1 and ends at Ala 2 (Ala1-Ala2 \[2xThr\]). The salt-dependent assembly of Ala1-Ala2 (2xThr) is more efficient than that of AD-Ala2 (2xThr) ([Figure 7](#pbio-0020356-g007){ref-type="fig"}E) at 5 μM protein. Thus, Ala 1 can partially overcome the inhibitory effect of Ala 2, but not as robustly as the C-terminal domain. A tail fragment containing both Ala 1 and the C-terminal domain (Ala1-Cterm \[3xThr\]) assembles essentially the same as AD-Cterm (3xThr) ([Figure 7](#pbio-0020356-g007){ref-type="fig"}E). All of these data indicate that the assembly reaction is delicately balanced, because the various domains of the tail make both favorable and unfavorable contributions to assembly. A Globular Head Need Not Be Located N-Terminal to the AD to Promote Assembly of Fixed-Size Structures {#s2k} ----------------------------------------------------------------------------------------------------- The 196-a.a. repeat is remarkably symmetric in the entire AD-Cterm region ([Figure 8](#pbio-0020356-g008){ref-type="fig"}A). We created a protein identical to GFP-AD-Cterm (3xThr) except that the GFP is located at the C-terminus, after a.a. 2,116 (AD-Cterm-GFP \[3xThr\]) ([Figure 8](#pbio-0020356-g008){ref-type="fig"}B). Because of the symmetry in this region of the myosin tail, the AD-Cterm-GFP (3xThr) and GFP-AD-Cterm (3xThr) proteins can distinguish between assembly mechanisms that require specific a.a. in specific positions relative to the globular head versus assembly mechanisms that require a very general charge pattern to occur. AD-Cterm-GFP (3xAsp) was also created to test whether the same three threonines can regulate assembly of this myosin tail fragment. ::: {#pbio-0020356-g008 .fig} Figure 8 ::: {.caption} ###### Design and Assembly Characteristics of AD-Cterm-GFP \(A) The AD-Cterm charge distribution (top) is aligned with the reverse charge distribution (bottom), showing the overall symmetry of the 196-a.a. charge repeat in this region of the tail. \(B) Analysis of assembly by EM. The AD-Cterm (3xThr) tail fragment has GFP on the C-terminus (AD-Cterm-GFP \[3xThr\]). The scale bar indicates a distance of 100 nm. Shown are three images of AD-Cterm GFP (3xThr), assembled 2--5 min. \(C) Analysis of assembly by sedimentation. The solubility of AD-Cterm-GFP (3xThr) and AD-Cterm-GFP (3xAsp) tail fragments constructs are compared to GFP-AD-Cterm (3xThr) and "headless" AD-Cterm (3xThr) tail fragments. ::: ![](pbio.0020356.g008) ::: AD-Cterm GFP (3xThr) Assembles into Bipolar Structures {#s2l} ------------------------------------------------------ To test whether AD-Cterm-GFP (3xThr) can form BTFs, we performed EM ([Figure 8](#pbio-0020356-g008){ref-type="fig"}B) and sedimentation assays ([Figure 8](#pbio-0020356-g008){ref-type="fig"}C). Similar to GFP-AD-Cterm (3xThr), AD-Cterm-GFP (3xThr) formed clustered or larger structures when protein was assembled for more than 10 min prior to imaging, whereas BTFs were most prevalent when protein was assembled for 2--5 min. The striations at the ends of the AD-Cterm-GFP (3xThr) bipolar structures are spaced 14 ± 2 nm (n = 118) apart, consistent with the 14-nm spacing of full-length Dictyostelium myosin. The AD-Cterm-GFP (3xThr) structures have a shorter bare zone than GFP-AD-Cterm (3xThr) structures (62 ± 8 \[n = 75\] versus 85 nm), but a similar width to both GFP-AD-Cterm (3xThr) and full-length myosin BTFs (32 ± 5 nm \[n = 159\] versus 27 nm and 33 nm, respectively). While the AD-Cterm-GFP (3xThr) structures differ slightly from those formed by full-length myosin and GFP-AD-Cterm (3xThr), it is clear that the presence of the globular domain leads to a fixed length, even in C-terminally capped myosin tail fragments. AD-Cterm GFP Assembly Is Regulated {#s2m} ---------------------------------- We next tested whether AD-Cterm-GFP assembly is regulated. AD-Cterm-GFP (3xAsp) is soluble at all NaCl concentrations, whereas AD-Cterm-GFP (3xThr) sediments efficiently between 25 mM and 150 mM NaCl ([Figure 8](#pbio-0020356-g008){ref-type="fig"}C), very similar to "headless" AD-Cterm and full-length myosin ([Figure 2](#pbio-0020356-g002){ref-type="fig"}A) as well as GFP-AD-Cterm (3xThr) ([Figure 3](#pbio-0020356-g003){ref-type="fig"}C). Therefore, regulation of assembly does not require a specific distance between the threonine phosphorylation sites and a globular head. Discussion {#s3} ========== Regulation of AD-Cterm and Full-Length Myosin Is Identical In Vitro {#s3a} ------------------------------------------------------------------- The assembly properties of both AD-Cterm (3xThr) and AD-Cterm (3xAsp) closely parallel those of full-length unphosphorylated and phosphorylated myosin described by [@pbio-0020356-Cote1]. These data represent a good basis for comparison to AD-Cterm, because the myosin was heavily phosphorylated (2:1 stoichiometry of phosphate:myosin heavy chain), and Cote and McCrea performed a gel filtration step to eliminate any contaminating actin. Other sedimentation data for full-length myosin compare favorably with the AD-Cterm data as well ([@pbio-0020356-Kuczmarski1]; [@pbio-0020356-Egelhoff1]). The parallel between the solubility of AD-Cterm and full-length myosin in vitro argues that AD-Cterm (3xAsp) is a good mimic of the phosphorylated state of full-length myosin and that all of the information necessary for regulated assembly is present in the AD-Cterm region of the myosin tail. Attaching a Globular Head to AD-Cterm Reconstitutes Regulated BTF Assembly {#s3b} -------------------------------------------------------------------------- The fact that GFP-AD-Cterm (3xThr) forms BTFs whereas AD-Cterm (3xThr) does not, demonstrates that the mere presence of a globular head differentiates BTFs from paracrystals, and that neither the myosin head nor the S2 region of the myosin coiled-coil is critical for BTF formation ([@pbio-0020356-Zang1]; this study). While the presence of the globular head is critical for the size and shape of myosin structures in this study, it has no appreciable effect on the regulation of myosin BTF assembly. Furthermore, other cellular factors, with the exception of kinases and phosphatases, are not required for BTF formation, because the process can be completely reconstituted using recombinant, bacterially expressed protein. Regulation of GFP-AD-Cterm and Full-Length Myosin Is Similar In Vivo {#s3c} -------------------------------------------------------------------- GFP-RLC-AD-Cterm tail fragments demonstrate regulated recruitment to the cleavage furrow during cytokinesis. While the localization of GFP-RLC-AD-Cterm (3xThr) is less robust than full-length myosin, constitutively assembled GFP-RLC-AD-Cterm (3xAla) always localizes to the cleavage furrow in cytokinesis (see [Figure 4](#pbio-0020356-g004){ref-type="fig"}). In contrast, GFP-RLC-AD-Cterm (3xAsp) fails to localize (see [Figure 4](#pbio-0020356-g004){ref-type="fig"}). These data are consistent with other observations ([@pbio-0020356-Sabry1]; [@pbio-0020356-Shu1]) indicating that assembled myosin constructs localize to the cleavage furrow while unassembled myosin constructs do not. The difference in localization efficiency between full-length myosin and GFP-RLC-AD-Cterm (3xThr) might be due to differences in degree of assembly. The critical concentration of full-length myosin is estimated to be less than 20 nM ([@pbio-0020356-Mahajan1]). One might expect the critical concentration of assembly for GFP-RLC-AD-Cterm (3xThr) to be higher than full-length myosin, because full-length myosin has a longer tail. Likewise, the critical concentration of GFP-RLC-AD (1xThr) is expected to be higher than that of GFP-RLC-AD-Cterm (3xThr). This argument may explain why GFP-RLC-AD-Cterm (3xThr) recruitment is less efficient than full-length myosin, and why GFP-RLC-AD (1xThr) does not go to the furrow at all (unpublished data). Interestingly, beads that are approximately the same length as the BTFs formed from full-length myosin do not localize to the cleavage furrow. Together with the previous in vivo localization data, this result argues that the majority of the information necessary for localization of BTFs to the cleavage furrow is in the AD-Cterm portion of the myosin tail, and that localization to the cleavage furrow is an active process, possibly involving another cellular factor that recruits myosin. It may be that this factor recognizes the charge repeats that distinguish assembled myosin BTFs from unassembled myosin. Consistent with this model is the localization of chimeric myosin molecules to the cleavage furrow in Dictyostelium cells ([@pbio-0020356-Shu2]; [@pbio-0020356-Shu3]). These chimeras have myosin tails from other species with no sequence homology to the Dictyostelium tail, but contain the charge repeats common to all myosin tails. Regulation Does Not Require an Ala 1-Ala 2 Intramolecular Interaction {#s3d} --------------------------------------------------------------------- Our finding that the in vitro self-assembly of AD-Cterm (3xThr) and AD-Cterm (3xAsp) are very similar to full-length, wild-type myosin and full-length, phosphorylated myosin, respectively, show that Ala 1 is not required for regulated assembly. This result has been confirmed in vivo by deleting Ala 1 from both full-length wild-type and 3xAsp myosin (W. Liang and JAS, unpublished data). Dictyostelium cells expressing the Ala 1 deletions as their sole source of myosin were phenotypically indistinguishable from cells expressing their full-length counterparts. It is possible that an Ala 1-Ala 2 intramolecular interaction is indeed one mode of regulation, but is not necessary, because redundant regulatory mechanisms occur at multiple points along the assembly pathway. Regulation Does Not Occur by Conformational Disruption of the Coiled-Coil {#s3e} ------------------------------------------------------------------------- The CD and analytical ultracentrifugation data show that AD-Cterm (3xAsp) has not simply failed to fold into a two-stranded α-helical coiled-coil. These data are consistent with rotary shadowed electron micrographs of full-length 3xAsp and phosphorylated myosin. Gross destabilization of the coiled-coil would result in a single-headed myosin, and none have been observed ([@pbio-0020356-Pasternak1]; [@pbio-0020356-Liang1]). Global destabilization might result in a section of the tail unfolding, lowering resistance to thermal denaturation. However, thermal melts show that the melting temperature of AD-Cterm (3xThr) and AD-Cterm (3xAsp) are identical (see [Figure 2](#pbio-0020356-g002){ref-type="fig"}C). Local destabilization of the coiled-coil seems unlikely, because one of the chymotrypsin cleavage sites is close to threonine 1,823, which is in a core position of the heptad repeat, yet the proteolytic susceptibility of the myosin tail does not change upon phosphorylation ([@pbio-0020356-Cote1]). Small numbers of assembled AD-Cterm (3xAsp) paracrystals were seen in electron micrographs of samples (unpublished data). These paracrystals, although quite rare, possess a 14-nm periodicity similar to the wild-type tail fragment. Consistent with this observation, approximately 10% of 3xAsp myosin sediments at 50 mM NaCl, demonstrating that regulation of assembly is not an all-or-none process. Together with the CD data, this result indicates that AD-Cterm (3xAsp) is structurally and functionally intact. Rather than being conformationally disrupted, the 3xAsp tail fragment fails to assemble because the critical concentration for assembly is higher than that of the wild-type tail fragment. Regulation of Assembly by Modulation of Charge-Charge Interactions {#s3f} ------------------------------------------------------------------ ### The role of the 196 a.a. charge repeat in assembly {#s3f1} The fact that AD-Cterm-GFP forms BTF-like structures with heads on the outside and tails in the center demonstrates the role of the 196-a.a. charge repeat in assembly. This charge repeat is symmetric in the AD-Cterm region of the tail (see [Figure 8](#pbio-0020356-g008){ref-type="fig"}A). If the 196 a.a. charge repeat is a major driving force of assembly, then BTF formation should be independent of whether a globular head is positioned on the N or C terminus of AD-Cterm. Since globular heads are clearly visible on the outer edges and not in the center of AD-Cterm-GFP (3xThr) structures, local interactions between a.a. in these filaments must be different than local interactions in filaments formed from GFP-AD-Cterm (3xThr) and from full-length myosin. Further structural characterization is required for a model that describes the packing of the coiled-coils within these filaments, but these data are consistent with the overall large scale charge character of the tail being important for forming BTFs. Interestingly, AD-Cterm-GFP (3xThr) structures are shorter than GFP-AD-Cterm (3xThr) filaments, and the bare zones are smaller as well. This may be because C-terminal sequence elements in the myosin tail may be required for proper formation of the bare zone and thereby determine the exact morphology of the BTF. ### The assembly reaction is delicately balanced. {#s3f2} The truncation analysis provides evidence of a delicate balance between forces that drive assembly and disassembly of myosin tails. This is most strikingly demonstrated by the observation that Ala 2 was actually found to inhibit the assembly of AD, while Ala 1 and the C-terminal domain help to drive assembly. We propose that that this delicate balance is related to the 196-a.a. charge repeat in the tail. The attachment of Ala 2 (purple, [Figure 1](#pbio-0020356-g001){ref-type="fig"}) to the AD (blue, [Figure 1](#pbio-0020356-g001){ref-type="fig"}) adds a large cluster of negative charge onto the end of this tail fragment. Assembly might be inhibited because a proper balance of charge is required for efficient self-assembly. This balance is restored by addition of either the C-terminal portion of the tail (green, [Figure 1](#pbio-0020356-g001){ref-type="fig"}) or Ala 1 (red, [Figure 1](#pbio-0020356-g001){ref-type="fig"}), because both of these domains possess clusters of positive charge. Because the charge repeats likely play a role in each step in the assembly pathway, a small effect, such as the introduction of negative charge at key points in the molecule, could have a large effect on overall self-assembly. The threonine phosphorylation sites are positioned near the positive clusters of charge at the end of AD and in the C-terminal domain, suggesting that regulation by phosphorylation might have its largest effect on this charge pattern. Materials and Methods {#s4} ===================== {#s4a} ### Construction of myosin tail fragments for expression in E. coli {#s4a1} The polymerase chain reaction (PCR) was used to amplify the regions of a Dictyostelium expression vector (pBIG) containing either GFP (3xThr) myosin or GFP (3xAsp) myosin ([@pbio-0020356-Sabry1]). An NdeI site was engineered at the 5′ end, and a stop codon followed by a SacI site was placed at the 3′ end of the PCR product. PCR products were directionally subcloned into the pET21a vector (Novagen, Madison, Wisconsin, United States) using these two restriction sites. Tail fragments contain an N-terminal methionine followed by these a.a. from Dictyostelium myosin: AD (1xThr and 1xAsp), a.a. 1,531--1,824; AD-Cterm (3xThr and 3xAsp), a.a. 1,531--2,116; extended AD (2xThr and 2xAsp), a.a. 1,531--1,840; AD-Ala 2 (2xThr and 2xAsp), a.a. 1,531--1,966; AD-2015 (2xThr), a.a. 1,531--2,015; Ala 1-Ala 2 (2xThr), a.a. 1,348--1,966; Ala 1-Cterm (3xThr), a.a. 1,348--2,116 (see [Figure 6](#pbio-0020356-g006){ref-type="fig"}). The same set of PCR primers were used to construct the wild-type and aspartic acid variants of each tail fragment except for the AD (1xAsp) fragment. For AD (1xAsp), Quikchange site-directed mutagenesis (Stratagene, La Jolla, California, United States) was used to mutate threonine 1,823 to aspartic acid. The sequence of all constructs was confirmed. ### Construction of GFP tail fragments for expression in E. coli {#s4a2} The constructs described above are the source of tail fragment DNA for all GFP constructions. To create N-terminal fusions, 5′ and 3′ ends of GFP-UV ([@pbio-0020356-Crameri1]) were modified using PCR with primers that contain an NcoI site followed by a 6xhistidine tag at the 5′ end and an NdeI site at the 3′ end before the GFP-UV stop codon. Internal NdeI and NcoI sites were eliminated from the GFP-UV gene by Quikchange mutagenesis (Stratagene), and the modified GFP-UV gene was subcloned into the pET28a vector (Novagen) using NcoI and NdeI (GFP-UV-pET28a). Tail fragments were subcloned into GFP-UV-pET28a using NdeI and NotI. To create C-terminal GFP fusions, the 5′ and 3′ ends of GFP-UV in the GFP-UV-pET28a vector were modified by PCR. The 5′ primer eliminated the N-terminal 6xhistidine tag and introduced a SacI site. The 3′ primer eliminated an internal SacI site and introduced a 6xhistidine tag, followed by a stop codon, followed by a NotI site. Quikchange mutagenesis (Stratagene) was used to eliminate the stop codon immediately before the SacI site in the AD-Cterm tail fragments contained in pET21a. The modified GFP-UV was subcloned into this vector using SacI and NotI. All DNA sequences were verified. ### Construction of GFP tail fragments for localization studies in Dictyostelium {#s4a3} To make N-terminal GFP-tail constructs for expression in Dictyostelium, the myosin regulatory light chain binding site (RLCBS) sequence was ligated into plasmid pTX-GFP ([@pbio-0020356-Levi1]) to create pTX-GFP-RLCBS. The GFP sequence from pTX-GFP was subcloned into pET28a (Novagen) using NcoI and SacI. The NdeI site was removed from the GFP sequence using Quikchange mutagenesis (Stratagene) and then subcloned back into pTX-GFP to create pTX-GFPΔNde1. The RLCBS was amplified from pBigGFPRLC+ ([@pbio-0020356-Zang1]) using primers to add a 5′ SacI site and 3′ NdeI and XhoI sites and ligated into pTX-GFPΔNde1 using SacI and XhoI digestion. This vector is called pTX-GFP-RLCBS. Myosin tail fragments were subcloned into pTX-GFP-RLCBS using the NdeI and XhoI sites from the corresponding pET21a-tail fragment vectors. All DNA sequences were verified. ### Protein expression in E. coli {#s4a4} All tail fragment constructs were transformed into the BL21-CodonPlus (DE3)-RIL strain (Stratagene). LB media contained 34 μg/ml chloramphenicol and either 100 μg/ml kanamycin for pET28a or 100 μg/ml carbenicillin for pET21a. Cells were grown at 37 °C to an absorbance at 600 nm of approximately 0.6. Protein expression was induced by adding 1 mM IPTG and incubating for 1 h. Cells were harvested by centrifugation at 6,370 × g for 15 min. ### Purification of myosin tail fragments from E. coli {#s4a5} Harvested cells were resuspended in 10 mM Tris (pH 7.4), 1 mM EDTA, 1 mM DTT, 500 mM NaCl, 30% sucrose, and protease inhibitor (PI) cocktail (final concentration of 0.7 μg/ml leupeptin, 0.7 μg/ml pepstatin A, 2 μg/ml aprotinin, and 1 mM PMSF). For a single protein prep, the pellet from 10 l of culture was resuspended in buffer for a final volume of 40 ml. Cells were added dropwise into liquid nitrogen and stored at --80 °C. Cells were thawed and 5 μg/ml RNase A (\#78020Y; USB, Cleveland, Ohio, United States), 50 μg/ml RNase-free DNase I (\#776 785; Roche, Basel, Switzerland), and 10 mM MgCl~2~ were added. The cells were lysed with two passes through a French Press (American Instrument Company, Silver Spring, Maryland, United States) at 10,000 pounds per square inch. After the first pass, a new batch of the PI cocktail was added. The lysate was centrifuged at 100,000 × g for 30 min at 4 °C. The supernatant was boiled for 10 min, and then a new batch of the PI cocktail and 1 mM DTT was added. The boiled supernatant was centrifuged at 100,000 × g for 30 min at 4 °C and then dialyzed against DEAE Low-Salt Buffer (DEAE-LSB; 10 mM Tris \[pH 8.0\], 1 mM EDTA, 1 mM DTT, and PI cocktail). The protein was injected onto a column consisting of eight tandem HiTrap DEAE Sepharose Fast Flow columns (Amersham Biosciences, Little Chalfont, United Kingdom). The column was washed with 5 volumes of DEAE-LSB and protein was eluted on a linear gradient from 0 to 500 mM NaCl over 10 volumes. Fractions containing protein (typically eluting at 230 mM NaCl) were identified by SDS-PAGE. Protein was precipitated by addition of 85% (NH~4~)~2~SO~4~, stirred at 4 °C for 30 min, and centrifuged at 100,000 × g for 30 min at 4 °C. The pellet was resuspended in a minimal volume of gel filtration buffer (10 mM Tris \[pH 8.0\], 1 mM EDTA, 1 mM DTT, and 500 mM NaCl) and loaded on a HiLoad 26/60 Superdex 200 column (Amersham Biosciences). Fractions were analyzed by SDS-PAGE, and those containing pure tail fragments were pooled and dialyzed against Mono-Q Low Salt Buffer (Mono-Q LSB; 10 mM Tris \[pH 8\], 1 mM EDTA). Protein was injected onto a Mono Q HR 5/5 column (Amersham Biosciences) and a gradient from 0 to 1 M NaCl was run. Tail fragments typically eluted at 450 mM NaCl. Protein concentration was determined by measuring the absorbance at 280 nm in 6M guanidine hydrochloride. The extinction coefficient was calculated using the sequence of one strand of the coiled-coil as described in [@pbio-0020356-Gill1]. ### Purification of GFP tail fragments from E. coli {#s4a6} After harvesting, cells were resuspended in 50 mM sodium phosphate (pH 8.0) with 250 mM NaCl and containing PI cocktail. For a single protein prep, the pellet from 30 l of culture was resuspended in buffer for a final volume of 120 ml. Lysis and the first centrifugation step were identical to the myosin tail fragment protocol described above except that 5 mM MgCl~2~ was added along with the DNase I and RNase A. The supernatant was batch-bound to Ni-NTA resin (Qiagen, Valencia, California, United States) at 4 °C for 1 h. The resin was washed with 40 ml of 50 mM sodium phosphate (pH 8.0) with 250 mM NaCl (starting buffer) followed by 20 ml of starting buffer containing 12.5 mM imidazole. Protein was eluted from the column using approximately 5 ml of starting buffer with 500 mM imidazole. Fractions containing protein were loaded onto a Hi-Load 26/60 Superdex 200 prep grade gel filtration column (Amersham). Both GFP-AD-Cterm and AD-Cterm-GFP tail fragments elute between 120 ml and 150 ml. GFP-AD elutes between 152 ml and 174 ml. Protein was dialyzed against Mono-Q LSB overnight and then injected onto the Mono-Q HR 5/5 column (Amersham Biosciences) as described above. Protein elutes at about 450 mM NaCl. Protein was used immediately after purification for microscopy and sedimentation assays. ### Electron microscopy {#s4a7} Purified GFP-tail fragments were diluted to a final protein concentration of approximately 1 μM, a final MgCl~2~ concentration of 10 mM, and a final NaCl concentration of 50 mM. This solution was deposited onto glow-discharged 300-mesh carbon stabilized copper grids coated with formvar (\#01753-F; Ted Pella, Redding, California, United States). AD-Cterm and AD proteins were assembled for 2 h, while GFP-containing proteins were assembled for 2 min. A 1% uranyl acetate solution was applied before imaging on a JEOL 1230 transmission electron microscope (JEOL USA, Peabody, Massachusetts, United States). ### Sedimentation assembly assay {#s4a8} Purified protein was dialyzed overnight against 10 mM imidazole (pH = 7.5), 0.1 mM EDTA, and 1 mM DTT. 10 μM protein was added to an equal volume of 10 mM imidazole, 0.1 mM EDTA, 1 mM DTT, 2× mM NaCl. The final concentration of NaCl was × mM NaCl, where × = 0, 25, 50, 75, 100, 150, and 250 mM. Samples were incubated on ice for at least 30 min and then centrifuged at 132,000 × g for 15 min at 4 °C. Three replicates of each sample were made to ensure reproducibility. Samples of the supernatant and pellet were run on SDS-PAGE gels. The intensity of bands was quantified with an AlphaImager 2000 Documentation and Analysis System (Alpha Innotech Corporation, San Leandro, California, United States). ### Circular dichroism {#s4a9} An Aviv Circular Dichroism Spectrometer model 62A DS was used (Aviv Corporation, Acton, Massachusetts, United States). The protein was in 10 mM Tris (pH 7.4), 500 mM NaCl, 1 mM EDTA, 1 mM DTT. Wavelength scans were taken from 260 nm to 200 nm. Data were collected every 1 nm with a 1-nm bandwidth and averaging time of 10 s. The spectra presented are buffer subtracted. Thermal melts were taken from 4 °C to 60 °C. θ~222~ was monitored with a 1 nm bandwidth. The temperature was increased in 1 °C increments, and the sample was equilibrated for 2 min at the new temperature before data collection. An averaging time of 30 s was used. The pmt dc voltage used was 1.0 V for 2.5 μM protein and 0.55 V for 50 μM protein. Kaleidagraph (Synergy Software, Essex Junction, Vermont, United States) was used for all data analysis. Plots of fraction denatured versus temperature were constructed as described in [@pbio-0020356-Allen1]. ### Analytical ultracentrifugation {#s4a10} AD-Cterm (3xThr) was assembled by diluting the NaCl concentration to 50 mM. Assembled protein was recovered by centrifugation at 100,000 × g for 15 min. Pellets were resuspended in 10 mM Tris (pH 7.4), 0.1 mM EDTA, 500 mM NaCl (High-Salt Buffer) and centrifuged at 100,000 × g for 15 min to remove aggregates. The AD-Cterm (3xAsp) tail fragment was centrifuged at 100,000 × g for 15 min to remove aggregates. Both tail fragments were exchanged several times into High-Salt Buffer using an Amicon Ultra-15 30,000 Dalton cut-off ultra-filtration device (Millipore, Billerica, Massachusetts, United States). The High-Salt Buffer was the blank in all analytical ultracentrifugation experiments. After buffer exchange, the protein was centrifuged at 14,000 rpm for 10 min in a microcentrifuge to remove aggregates. The protein was snap-frozen in liquid nitrogen and shipped in dry ice to the Macromolecular Interactions Facility at UNC-Chapel Hill. The protein used for all analytical ultracentrifugation experiments was centrifuged at room temperature for 15 min at 16,000 × g to remove aggregates. All sedimentation equilibrium experiments were performed at the UNC Chapel Hill Macromolecular Interactions Facility as described in [@pbio-0020356-Patel1]. The rotor speed was set at 10,000 rpm, the temperature was maintained at 10 °C, and for the meniscus depletion experiment the conditions were centrifugation at 45,000 rpm for 8 h. AD-Cterm (3xThr) and AD-Cterm (3xAsp) were examined at protein concentrations of 52 μM, 35 μM, 17 μM, and 10.5 μM in High-Salt buffer. The measured densities of the buffers were 1.0 g/ml at 20 °C, and the partial specific volume of both tail fragments was calculated to be 0.73 ml/g at 10 °C ([@pbio-0020356-Durschschlag1]). All data analysis was done using XL-A/XL-I data analysis software version 4.0 (Beckman, Fullerton, California, United States). ### Culture of Dictyostelium cells expressing GFP-RLC-tail fragments {#s4a11} HS1 myosin-null Dictyostelium cells ([@pbio-0020356-Ruppel1]) were transformed using electroporation. Cells were grown in 10-cm petri dishes at 22 °C in HL5 medium ([@pbio-0020356-Sussman1]) supplemented with 60 μg/ml penicillin, 60 U/ml streptomycin, and 15 μg/ml G418 (Life Technologies, Carlsbad, California, United States) for plasmid selection. ### Sedimentation of lysates prepared from Dictyostelium cells expressing GFP-RLC-tail fragments {#s4a12} 1 ml of nearly confluent Dictyostelium cells expressing the appropriate construct was centrifuged and resuspended in 10 mM imidazole (pH 7.5), 0.1 mM EDTA, and 50 mM NaCl in the presence of protease inhibitors. The cells were lysed by freezing in liquid nitrogen and thawing. Cell lysates were allowed to sit on ice for 2 h before centrifugation at 132,000 × g for 20 min. The pellets were resuspended in equal volumes of the lysis buffer. The supernatant and pellets were resolved by SDS-PAGE without boiling the samples. The GFP-fusion proteins were imaged in the gel by excitation at 532 nm on a Typhoon 8000 imager (Molecular Dynamics, Sunnyvale, California, United States). ### Creation of Dictyostelium GFP-myosin expressing stable cell line {#s4a13} The GFP sequence was integrated into the Dictyostelium genome upstream of and in-frame with the gene encoding myosin-II (mhcA) by homologous recombination. The 167 bp just upstream of mhcA through the first 500 bp of the coding region were PCR-amplified from Dictyostelium genomic DNA using primers to add a 5′ XhoI site and a 3′ BamHI site. The PCR product was cloned into pBluescript digested with XhoI and BamHI. A PstI site was added between the upstream and mhcA coding regions by Quickchange mutagenesis (Stratagene) to create plasmid pBS/upstream-mhcA. GFP was amplified from pTX-GFP ([@pbio-0020356-Levi1]) with primers to add 5′ and 3′ PstI sites. The PCR product was then cloned into pBS/upstream-mhcA digested with PstI to create pBS/upstream-GFP-mhcA. The resulting plasmid was digested with ApaI and XbaI, and the fragment containing GFP flanked by the upstream and coding regions of mhcA was gel purified. The fragment was mixed with the blasticidin-resistant plasmid pBsr2 ([@pbio-0020356-Sutoh1]) in a 10:1 excess and transformed into Dictyostelium. After the transformed cells were selected with 4 μg/ml blasticidin S (ICN Biochemicals, Costa Mesa, California, United States), a clonal cell line, GMO8B, was created by plating FACS-sorted GFP-positive cells on Klebsiella lawns and picking spores from individual GFP-positive plaques. The correct integration of GFP upstream of mhcA was verified by PCR and imaging of GFP fluorescence in-gel after SDS-PAGE of cell lysates. The cell line was able to develop normally and grow in suspension, and GFP localization in GMO8B was very similar to GFP-myosin localization in cells expressing GFP-myosin from a plasmid ([@pbio-0020356-Moores1]). ### Scrape loading of beads into Dictyostelium cells {#s4a14} Glass slides were coated with 10 μg/ml polylysine overnight. Approximately 1 ml of semi-confluent GMO8B Dictyostelium cells in HL5 medium were allowed to attach to the polylysine-coated slide for 30 min. The medium was then replaced by a 0.04% solution of 0.5 μm-diameter carboxylated red FluoSpheres (Molecular Probes, Eugene, Oregon, United States) in HL5 medium. The cells were immediately scraped off of the surface with a rubber policeman and transferred to a clean glass slide. After the cells were allowed to attach for approximately 30 min, the slide was rinsed in PBS to remove excess beads not taken up by the cells. The cells were then removed from the glass slide by pipetting up and down with HL-5 medium. ### Live cell fluorescence microscopy {#s4a15} Live cells were transferred to imaging chambers (Applied Scientific, Santa Ana, California, United States) in HL-5 medium. GFP and FluoSphere fluorescence were imaged at room temperature using a Zeiss (Oberkochen, Germany) Axiovert 200 inverted epifluorescence microscope equipped with a 63× objective (N.A. 1.3). Cells were imaged at 20-s intervals. Images were collected using Metamorph (Universal Imaging Corporation, Downington, Pennsylvania, United States) and analyzed with ImageJ (NIH) and Photoshop (Adobe Systems, San Jose, California, United States). We would like to thank Ben Spink for helping us analyze charge distributions with Microsoft Excel, John Perrino and Nafisa Ghori for assistance with negative staining EM, the Harbury lab for help with CD, Gary Pielak for providing resources and support at UNC-CH, and the Spudich lab for critical reading of the manuscript. Special thanks to Hans Warrick for several thorough reads of the manuscript. This work was supported by NIH GM46551 (JAS) and American Heart Association 0435341Z (SER and PMM). Conflicts of interest. The authors have declared that no conflicts of interest exist. Author contributions. DRH, SER, and JAS conceived and designed the experiments. DRH, SER, SOD, DA, PMM, and SS performed the experiments. DRH, SER, and PMM analyzed the data. SOD and AT contributed reagents/materials/analysis tools. DRH and SER wrote the paper. Academic Editor: Manfred Schliwa, Adolf-Butenandt-Institut ¤1 Current address: Department of Pharmaceutical Chemistry, University of California, San Francisco, California, United States of America ¤2 Current address: Department of Cell and Molecular Biology, Northwestern University, Chicago, Illinois, United States of America Citation: Hostetter D, Rice S, Dean S, Altman D, McMahon PM, et al. (2004) Dictyostelium myosin bipolar thick filament formation: Importance of charge distribution and specific domains of the myosin rod. PLoS Biol 2(11): e356. 3xAla myosin : myosin II with three threonine heavy chain phosphorylation sites at positions 1,823 3xAsp myosin : myosin-II with three threonine heavy chain phosphorylation sites at positions 1,823 a.a. : amino acid(s) AD : the 34-kDa assembly domain (a.a. 1,531--1,824) AD-Cterm : the 68-kDa tail fragment starting at the AD and extending to the end of the C terminus of myosin-II (a.a. 1,531--2,116) BTF : bipolar thick filament CD : circular dichroism Cterm : C-terminal domain (a.a. 1,967--2,116) EM : negative staining electron microscopy GFP : green fluorescent protein RLC : regulatory light chain θ~222~ : mean residue ellipticity at 222 nm	PubMed Central	2024-06-05T03:55:48.563743	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523230/", "journal": "PLoS Biol. 2004 Nov 19; 2(11):e356", "authors": [ { "first": "Daniel", "last": "Hostetter" }, { "first": "Sarah", "last": "Rice" }, { "first": "Sara", "last": "Dean" }, { "first": "David", "last": "Altman" }, { "first": "Peggy M", "last": "McMahon" }, { "first": "Shirley", "last": "Sutton" }, { "first": "Ashutosh", "last": "Tripathy" }, { "first": "James A", "last": "Spudich" } ] }
PMC523231	Understanding how the brain processes and stores information depends in large part on knowing which neurons are involved in a particular process and how they\'re organized into functional networks. Each of the 10 billion or so neurons in the brain has thousands of connections to other neurons, sending (via axons) or receiving (via dendrites) the signals that allow us to think. Each neuron can transmit signals to both local and distant neurons, and it is by mapping these networks that neuroscientists can discern correlations between neural connectivity and physiological responses and ultimately unveil the computational algorithms underlying brain function. Since the beginning of cellular neuroscience at the end of the 19th century, neuronal connections have been explored by tracing axons and dendrites under the light microscope. But even with the resolution of state-of-the-art light microscopy, this approach works only if a small subset of neurons is stained and thus leaves most of the network hidden. Electron microscopy, on the other hand, can provide the spatial resolution necessary both to resolve processes in densely packed neural "wire bundles" and to identify synapses faithfully, but individual electron microscopic images are restricted to two dimensions. Transmission electron microscopy provides cross-sectional images through tissue, while scanning electron microscopy typically provides the appearance of 3D but in reality maps only the specimen surface and is thus blind to the connections within. It\'s possible to wrest 3D information from the transmission electron microscope by using tilt-series tomography, but sections can\'t be much thicker than 1 micron (a millionth of a meter). Data from thicker volumes can be obtained, but the process so far has been so painstaking and time-intensive---it involves, among other labor-intensive tasks, manually reconstructing serial sections---that few undertake it. It should, however, be possible to get similar data with "serial block-face imaging," which involves repeatedly cutting section after section from a plastic-embedded block of tissue and photographing what\'s left. Scanning electron microscopy is needed for this task, but sample preparation methods are like those used for transmission electron microscopy, albeit with a few additional steps to enhance contrast. ::: {#pbio-0020388-g001 .fig} ::: {.caption} ###### Neurite Reconstruction Manual reconstruction of selected processes in cortical tissue ::: ![](pbio.0020388.g001) ::: This is exactly what Winfried Denk and Heinz Horstmann have done to obtain "truly 3D datasets" using a method they call "serial block-face scanning electron microscopy" (SBFSEM), for which they constructed a "microtome" that goes inside the scanning electron microscope chamber. The resolution achieved is sufficient to reveal "even the thinnest of axons" and identify synapses. The SBFSEM method can generate stacks of thousands of ultra-thin sections, 50--70 nanometers (a nanometer is a billionth of a meter) thick, generating 3D datasets to reconstruct the topology and circuitry of neurons in brain tissue. The authors\' custom-designed microtome holds the tissue block in a way that ensures image alignment and maintains focus; all the while the specimen surface is positioned close enough to the objective lens to allow high-resolution imaging. Denk and Horstmann expect that with this method they might ultimately be able to cut sections thinner than the 50 nanometers that their current setup manages. This then would allow them to cut sections even thinner than what is routinely possible in conventional transmission electron microscopy. While the authors doubt that the lateral resolution will ever reach that of transmission electron microscopy, they also argue that such high resolution may not actually be needed to trace neuronal connectivity. On the other hand, the method accelerates 3D electron microscopic data collection "by several orders of magnitude" by obviating the need for the labor-intensive adjustments to correct alignment and distortion required by other methods, an advance that is crucial for large-volume neuroanatomy and might, in addition, open up many hitherto inaccessible problems to ultra-structural investigations.	PubMed Central	2024-06-05T03:55:48.568925	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523231/", "journal": "PLoS Biol. 2004 Nov 19; 2(11):e388", "authors": [] }
PMC523232	The movements needed to read this synopsis---turning the page, tracking along the lines, sitting, breathing---all require myosin, a molecular motor in muscle that transforms chemical energy into small but deliberate motions. But beyond these macro-movements, myosin is also required for the micro-movements of individual cells and their organelles and for determining cellular architecture. There are many different myosins, but they all have the same general structure. At one end is a globular head, which is responsible for motor activity. This head binds ATP---the cell\'s power supply---and actin, an important component of the cytoskeleton of cells. Next comes a helical neck or lever region. Finally, there is a long helical tail, which has different and somewhat poorly understood functions in the different myosins. Myosin II, the classical form of myosin found in essentially all eukaryotic cells, is constructed from two heavy chains (which contain the three regions described above) and two pairs of light chains (which stabilize the neck region). The long helical tail of myosin II is formed by the two heavy chains wrapping around each other and is involved in getting myosin II to the right place in the cell, as well as in assembling it into filaments. Individual myosin II molecules can make tiny molecular motions. ATP cleavage induces a shape change in the globular head, which is transmitted to the lever region of the molecule. Angular rotation of this region moves the myosin along the actin filament. But to achieve the larger movements that are necessary to, for example, split cells apart during cell division, individual myosin II molecules group together to form highly regular bipolar structures called bipolar thick filaments (BTFs). In these, the globular myosin heads are positioned on either side of the filament, and the tail regions are clustered in the middle. This geometry enables myosin II molecules in thick filaments to pull from either side, generating contractile forces. James Spudich\'s team has been studying the assembly of these thick filaments in the slime mold Dictyostelium discoideum, an organism beloved by developmental and cellular biologists because of its simple development and ease of manipulation. In the present study, the researchers examined the physical properties of various fragments of the myosin tail to find out how the self-assembly and disassembly of the BTFs are regulated. They already knew that the addition of phosphate groups on three specific threonine amino acid residues in this region (through a chemical reaction called phosphorylation) is important for regulating BTF assembly; they knew this from studies showing that mutation of these residues to aspartic acid, which mimics phosphorylated threonine, inhibits BTF formation. Here, the researchers show that a specific tail fragment of the myosin heavy chain containing the three crucial threonine residues assembles into a structure with some, but not all, of the properties of BTFs. However, replacing these threonine residues with aspartic acid prevents any self-assembly of the fragment. Further experiments in which different tail regions were nibbled away and the assembly properties of the remaining fragments were determined suggest that the myosin tail contains a series of elements that correlate with the distribution of charged amino acids along the tail, some of which favor assembly and some of which favor disassembly. But it\'s not just the tail that is important. For myosin II to form fully fledged BTFs of a defined size, it seems that the addition of some kind of globular head---in these experiments one composed of green fluorescent protein so that it could be examined---is necessary. The overall result is a molecule that is finely poised to self-assemble into BTFs in response to one or two charge changes produced by phosphorylation. Consequently, the myosin contractile system can respond rapidly to environmental changes. Although Dictyostelium myosin II is somewhat different from vertebrate myosin II, the general principle by which myosin assembly and disassembly are regulated seems likely to hold for other myosins and so might throw light onto human disorders that involve myosin defects. But more fundamentally, similar principles may hold for spatial and temporal regulation of the many other macromolecular assemblies that are at the heart of cell and developmental biology.	PubMed Central	2024-06-05T03:55:48.569621	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523232/", "journal": "PLoS Biol. 2004 Nov 19; 2(11):e392", "authors": [] }
PMC523233	The alarm clock rings and you jump straight out of bed, rattling downstairs to start the day. Or maybe you creak downstairs, each step a struggle because of stiffness and pain in your knees and other joints. If the second description fits the start of your day, then maybe, like 70 million Americans, you have arthritis, one of the most prevalent chronic health problems in the United States. Arthritis is an umbrella term for more than 100 medical conditions. What all forms of arthritis have in common is that they affect our joints-places where two or more bones meet. In healthy joints, the ends of the bones are covered with cartilage, a tough but smooth tissue that, like the oil in a car engine, reduces friction between the moving parts. In the most common form of arthritis-osteoarthritis-breakdown of this cartilage, which is called articular cartilage, means the bones rub together, causing pain and loss of movement. Risk factors for osteoarthritis include age and family history. If we could understand the molecular mechanisms that create and maintain articular cartilage, it might be possible to discover what goes wrong in our joints as we age and to find better treatments for arthritis. Embryologists have already discovered quite a bit about the earliest stages of joint formation. It is known, for example, that stripes of cells that form between developing bones subsequently develop into the permanent cartilage found in joints. Several members of a family of secreted proteins known as bone morphogenetic proteins (BMPs) are expressed in these stripes of cells, implicating BMP signaling (the transmission of messages produced by BMPs binding to cell-surface receptors) in early joint development. ::: {#pbio-0020395-g001 .fig} ::: {.caption} ###### Targeting genes in joints ::: ![](pbio.0020395.g001) ::: David Kingsley\'s team has been investigating whether BMPs are also involved in the later development and maintenance of joint cartilage. To do this, the researchers designed a genetic system that inactivates BMP signaling late in mouse embryonic development. They inserted special DNA sequences called loxP sites on either side of Bmpr1a, a gene that encodes one of the BMP receptors. The loxP sites have no effect until an enzyme known as Cre is expressed, and then the DNA between the loxP sites is cut out and discarded. Because Kingsley\'s team knew that global inactivation of Bmpr1a early in development causes embryonic death, they linked the gene for Cre to DNA sequences that limit its expression to those regions of the embryo where joints eventually develop. The result: a mouse strain in which Bmpr1a receptor function is specifically lost only in tissues destined to become joints. Most of the joints in this mouse strain formed normally. However, the mice rapidly developed severe arthritis after birth. By 7 days old, the expression of proteins normally found in cartilage was reduced, although at this stage the knee, for example, looked normal. By 7 weeks old (adulthood for mice), there were clear structural changes in the knee joints, and the articular cartilage was thinner and showed signs of wearing away. By 9 months old, the knees of the mutant mice largely lacked articular cartilage and the unprotected leg bones seemed to rub directly against each other. All told, the joints in these mutant mice closely resembled those in people with osteoarthritis, suggesting that BMP signaling is necessary for the maintenance of healthy articular cartilage. This raises the possibility that mutations in BMP signaling components may underlie some of the genetic variation in human osteoarthritis risk and suggests that treatments designed to mimic or augment BMP signaling might help to maintain healthy joints. Finally, the genetic system described by Kingsley and coworkers should be useful for future investigations into joint formation and maintenance.	PubMed Central	2024-06-05T03:55:48.570210	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523233/", "journal": "PLoS Biol. 2004 Nov 19; 2(11):e395", "authors": [] }
PMC523234	For more than 2,000 years, from the time of Aristotle onwards, it was thought that the complete body plan of human beings (and that of other animals) was present in the fertilized egg. During pregnancy, a preformed miniature human being, or homunculus, grew bigger and bigger; development was simply a process of growth. Then, in the mid-18th century, Carl Friedrich Wolff described how the chick gut, basically a tube, forms from an initially flat sheet of cells, overthrowing at a stroke the preformation theory of embryology. We now know that development is a complex series of coordinated processes that transforms the amorphous ball of cells produced from the fertilized egg by cell division into an intricate body containing numerous specialized tissues and organs. And we are beginning to understand how a wide array of transcription factors---proteins that bind to regulatory sequences within genes to control their expression---guide the sequential stages involved in development. It seems that these factors form regulatory networks that control the temporal and spatial waves of gene expression that underlie and are required for organized body building. Susan Mango and her colleagues are studying the role of transcription factors in controlling organ development. The organ they are studying---the pharynx of the nematode worm---is relatively simple. This muscular tube, which passes bacteria (the food of this small soil-dwelling organism) from the mouth to the midgut, contains fewer than 100 cells of only seven different types. To get an overall picture of the regulatory sequences within genes that are involved in the temporal control of pharyngeal development, the researchers identified 339 candidate pharyngeal genes by comparing gene expression profiles in mutant worm embryos that had excess pharyngeal cells with those in mutant embryos lacking pharyngeal cells. Then, by referring to a database that details gene expression patterns in nematode worms and embryos, the researchers classified 37 of their candidate genes as having early-onset expression and 34 as having late-onset expression. Next, the scientists carefully examined the DNA sequence of each gene for candidate regulatory regions that might contribute to its temporal regulation. Of nine candidate motifs revealed by this search, six functioned as regulatory sites in in vivo assays. The researchers estimated that these six elements, together with sites that bind PHA-4---a member of a family of transcription factors that are important in digestive tract development in many animals---account for the timing of onset of expression of about half of the nematode\'s pharyngeal genes. Finally, the researchers used combinations of the newly discovered temporal regulation sites and PHA-4 sites in a genome-wide search that predicted pharyngeal genes and their time of onset of expression with greater than 85% accuracy. ::: {#pbio-0020409-g001 .fig} ::: {.caption} ###### Fluorescent reporter genes expressed in the developing C. elegans foregut ::: ![](pbio.0020409.g001) ::: From these results and those of previous studies, Mango and her colleagues propose a model to explain how the temporal control of pharyngeal gene expression needed for pharynx development is achieved. The earliest time for pharyngeal gene expression, they suggest, is determined by how well PHA-4 sticks to a particular gene\'s binding site. However, gene expression only occurs if other factors that bind to the regulatory sites are also present, and the exact combination of these factors determines which gene is active at any given time. The identity of these factors remains to be discovered. Nevertheless, at least for this simple organ, we now have a much better idea of how the complex process of organ formation is controlled at a molecular level, and it is likely that similar regulatory networks will underlie the formation of other organs as well.	PubMed Central	2024-06-05T03:55:48.570900	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523234/", "journal": "PLoS Biol. 2004 Nov 19; 2(11):e409", "authors": [] }
PMC523824	Introduction {#s1} ============ Celiac disease is a chronic inflammatory condition caused by an inappropriate immune response of intestinal T-cells reactive to gluten proteins of wheat and similar prolamin proteins of related cereals \[[@pmed-0010001-b01]\]. The majority of the peptides recognized by intestinal T-cells are more immunogenic following deamidation by tissue transglutaminase (TG2). These peptides are invariably presented by HLA-DQ2 or -DQ8, the same HLA molecules that confer genetic predisposition to celiac disease \[[@pmed-0010001-b01]\]. Gluten-reactive intestinal T-cells can be isolated from virtually all patients with celiac disease but not from normal individuals. The disease goes into remission when harmful cereals are avoided. A gluten-free diet is thus the standard treatment of this disorder. Oats have traditionally been excluded from the gluten-free diet. Several feeding studies, however, have indicated that patients with celiac disease and dermatitis herpetiformis tolerate oats without signs of intestinal inflammation \[[@pmed-0010001-b02]--[@pmed-0010001-b09]\]. Of note, some of these studies have high patient-dropout rates that may have masked cases of oat intolerance. An in vitro study found no signs of T-cell activation in small intestinal biopsies of celiac disease patients challenged with avenin (the prolamin fraction of oats) \[[@pmed-0010001-b10]\], and avenins have been predicted to contain only a few glutamines that can be deamidated by TG2, presumably making avenins less immunogenic \[[@pmed-0010001-b11],[@pmed-0010001-b12]\]. On this basis, oats have been allowed in the gluten-free diet in several countries \[[@pmed-0010001-b13]\]. It remains to be proven that all celiac disease patients tolerate oats following long-term exposure. A recent study of 39 Finnish patients randomized to eat a gluten-free diet with 50 g of oats daily or a standard gluten-free diet for 1 y reported more intestinal symptoms and more gut inflammation in the group of patients eating oats, although the mucosal integrity was not disturbed \[[@pmed-0010001-b14]\]. In an open challenge study of 19 adult celiac disease patients using pure oats, one patient developed villous atrophy \[[@pmed-0010001-b15]\]. This finding prompted us to investigate the phenomenon of oat intolerance further in a selected series of nine adult celiac disease patients, three of whom had clinical oat intolerance. The goal of the study was to characterize the intestinal T-cell response to oats avenin proteins in these patients in detail and to relate this to clinical symptoms and intestinal biopsy findings. Methods {#s2} ======= Participants {#s2a} ------------ We studied nine adults with celiac disease who had a history of exposure to pure oats. The oats were derived from a quality-controlled production line and were shown to be free from contamination of other cereals as described elsewhere \[[@pmed-0010001-b15]\]. The selection of the study participants was not random. Five of the patients (CD359, CD377, CD422, CD431, and CD482) participated in a clinical challenge study consisting of 19 adults with celiac disease who ate 50 g oats daily for 12 wk \[[@pmed-0010001-b15]\]. One of these patients (CD422) has symptoms and mucosal inflammation on oats consumption as described \[[@pmed-0010001-b15]\]. Patient CD431 has slight mucosal inflammation when eating oats but is clinically well. The three remaining individuals eat and tolerate oats. All these five patients agreed to undergo gastroduodenoscopy for research purposes. In addition, two other adults with celiac disease (CD446 and CD504) were recruited from our ordinary outpatient clinic. Patient CD446 eats and tolerates oats, whereas patient CD504 has anaphylactoid symptoms after intake of oats but has no mucosal inflammation. Finally, two patients (CD496 and CD507) were referred by a general practitioner and a referring hospital for investigation of complications arising when eating a gluten-free diet, here termed complicated celiac disease. The latter four patients came for gastroduodenoscopy for clinical reasons, and agreed to have extra biopsies taken for research purposes. We were unable to measure serological parameters in these last four patients because no serum samples were taken from them during their clinical course. The study was approved by the regional ethical committee. The participants gave their informed consent. Histopathological Assessment {#s2b} ---------------------------- We took small intestinal biopsies from the horizontal part of the duodenum by gastroduodenoscopy using an Olympus (Tokyo, Japan) GIF-IT140 scope and scored them according to the modified Marsh criteria \[[@pmed-0010001-b16]\]. Intraepithelial lymphocytes were counted in hematoxilin-eosin-stained sections and enumerated per 100 enterocytes. Five areas per biopsy were counted, each encompassing 50--100 epithelial cells. Grain Antigens and Peptides {#s2c} --------------------------- Oat grains (Regal, Oslo, Norway) were ground and the flour was washed twice with water-saturated 1-butanol. The pellet was dissolved in 45% ethanol overnight and centrifuged. The avenin fraction was precipitated from the supernatant by adding two volumes of 1.5 M NaCl. The precipitate was dissolved either in 0.01 M acetic acid (pH 1.8) and digested with pepsin and subsequently trypsin (pH 7.8) or in 2 M urea /0.01 M NH~4~HCO~3~ and digested with chymotrypsin. Gluten and gliadin (ethanol-soluble proteins of gluten) were isolated from household wheat flour and digested with chymotrypsin as described \[[@pmed-0010001-b17]\]. Avenin peptides were synthesized on a robotic system (Syro MultiSynTech, Bochum, Germany) using Fmoc/OtBu chemistry and 2-chlorotrityl resin (Senn Chemicals, Dielsdorf, Switzerland). The identity of the peptides was confirmed by electrospray mass spectrometry, and purity was analyzed by reverse-phase HPLC. Treatment of the peptides with guinea pig (Sigma; St. Louis, Missouri, United States) and human recombinant TG2 was performed in the presence of 1 and 2 mM CaCl~2~, respectively, at 37 °C for 2 h. T-Cell Assays {#s2d} ------------- The generation of T-cell lines, T-cell cloning, and T-cell proliferation assays were performed as described elsewhere \[[@pmed-0010001-b17]\]. Single biopsy specimens from the patients were challenged in vitro with either a pepsin-trypsin digest or a chymotrypsin digest of avenin. As control, single biopsy specimens were challenged with a chymotrypsin digest of gluten or gliadin. DR3+DQ2+ B lymphoblastoid cells (irradiated 80 Gy) were used as antigen-presenting cells. Positive T-cell responses were defined as a stimulatory index (SI) (\[T + APC + antigen\] divided by \[T + APC\]) above 3. Determination of HLA restriction of the T-cells was done by testing inhibition of T-cell proliferation by the monoclonal antibodies B8.11 (anti-DR), SPV-L3 (anti-DQ), 2.12.E11 (anti-DQ2), and W6/32 (anti-HLA class I). Purification of Avenin Fragments {#s2e} -------------------------------- A pepsin-trypsin digest of avenin was separated by gel filtration (FPLC, column Superdex Peptide HR 10/30; Pharmacia Biotech, Uppsala, Sweden), and a fraction containing T-cell stimulatory fragments was further separated by reverse-phase HPLC (Äkta, Pharmacia Biotech; column Jupiter 5μ C18, 250 × 4.6 mm, Phenomenex, Torrance, California, United States) using an acetonitrile gradient from 5% to 40% with 1%/min and from 40% to 64% with 3%/min (flow rate 0.9 ml/min, containing 0.05% trifluoroacetic acid). Mass Spectrometry and Database Searching {#s2f} ---------------------------------------- Electrospray ionization tandem mass spectrometry was performed on a quadrupole time-of-flight hybrid mass spectrometer (Micromass, Manchester, England). For spraying, needles were typically held at 900 V towards a skimmer cone (40 V). In collision-induced dissociation of selected peptide ions (collision gas argon; collision energy 25--35 eV), the generated characteristic b- and y-type fragment ions \[[@pmed-0010001-b18]\] were detected by the orthogonal TOF mass analyzer. All tandem mass spectrometry spectra were centroided and searched against in the NCBInr database via the Mascot Search Engine (<http://www.matrixscience.com>). Results {#s3} ======= Clinical and Histological Characteristics {#s3a} ----------------------------------------- Nine adults with celiac disease who had a history of exposure to oats assessed to be free from contamination of other cereals were studied. In some cases they came for gastroduodenoscopy for clinical reasons, in other cases, they agreed to come for research reasons. The characteristics of the patients are given in [Table 1](#pmed-0010001-t001){ref-type="table"}. This case series is thus not a consecutive series of ordinary patients with celiac disease. Three of these patients (CD422, CD496, and CD507) were known to exhibit clinical and histopathological signs of oat intolerance. Patient CD422 developed villous atrophy and dermatitis while eating oats, and details of this patient are described elsewhere \[[@pmed-0010001-b15]\]. Patient CD496 was a 53-y-old woman who was evaluated for complicated celiac disease. Celiac disease was diagnosed in 1987 after 1 y with diarrhea and weight loss; a biopsy showed a Marsh 3C lesion with an intraepithelial lymphocyte (IEL) count of 58/100 enterocytes (range 53--69) ([Figure 1](#pmed-0010001-g001){ref-type="fig"}). She responded well to a standard gluten-free diet. A control biopsy was not taken. In 2001, she started eating pure oats, but lost weight, going from 55 kg to 44 kg. While eating oats, a biopsy showed a Marsh 3A lesion with an IEL count of 54/100 enterocytes (range 43--62). The oats were discontinued, and she gradually recovered. Some months later, an intestinal biopsy demonstrated a Marsh 1 lesion with an IEL count of 46 (range 28--52). Clinically she is currently well. Patient CD507 was a 59-y-old woman who was also evaluated for complicated celiac disease. She probably had undiagnosed celiac disease since childhood and was diagnosed in 1990 after osteoporotic fractures. A biopsy showed total villous atrophy (Marsh 3C) with an IEL count of 50/100 enterocytes (range 44--54) ([Figure 1](#pmed-0010001-g001){ref-type="fig"}). She responded well to a standard gluten-free diet. In 1999, a follow-up biopsy showed complete normalization of her mucosa (Marsh 0) with an IEL count of 26/100 enterocytes (range 24--32). In 2000, the patient started eating oats and developed bloating, abdominal pain and iron deficiency. She lost 2 kg in weight. In 2002, while still eating oats, a biopsy showed a Marsh 3A lesion with an IEL count of 50/100 enterocytes (range 38--58). She discontinued eating oats and improved clinically. A new biopsy later in 2002 showed improvement, with a Marsh 1 lesion with an IEL count 32/100 enterocytes (range 24--46). Surprisingly, in late 2003 she was diagnosed with an adenocarcinoma in the small intestine, which was removed surgically. ::: {#pmed-0010001-g001 .fig} Figure 1 ::: {.caption} ###### Histology of Intestinal Mucosa of Two of the Oat-Intolerant Patients Small intestinal biopsies were obtained at diagnosis, after an ordinary gluten-free diet (remission), after introduction of oats, and after withdrawal of oats (recovery). For patient CD496, a biopsy was not taken after she started with a gluten-free diet. Biopsies were scored according to the modified Marsh criteria. Hematoxilin-eosin staining was used, and IEL counts are given in the corners of the photomicrographs. The remission biopsy from patient CD507 was poorly oriented. We therefore melted and reoriented this biopsy (insert). Original magnification: 100×. ::: ![](pmed.0010001.g001) ::: ::: {#pmed-0010001-t001 .table-wrap} Table 1 ::: {.caption} ###### Characteristics of the Included Patients ::: ![](pmed.0010001.t001) ::: Avenin-Reactive T-Cell Lines Generated from Intestinal Biopsies Challenged with Avenin {#s3b} -------------------------------------------------------------------------------------- Responses to TG2-treated avenin were detected in the polyclonal T-cell lines derived from the avenin-challenged biopsies from all three patients who had clinical and histopathological signs of oat intolerance ([Table 1](#pmed-0010001-t001){ref-type="table"}). Intestinal T-cell responses to TG2-treated avenin were also found in two of the other six patients. At least one avenin-reactive T-cell line from each patient was expanded. Inhibition experiments using anti-HLA class I and class II monoclonal antibodies demonstrated that these T-cell lines were all restricted to DQ2 (Figure 3; unpublished data), and with the exception of the T-cell line generated from patient CD482, they all gave an enhanced T-cell response to avenin treated with TG2 compared to avenin not treated with TG2 ([Table 2](#pmed-0010001-t002){ref-type="table"}). T-cell lines derived from the biopsies challenged ex vivo with avenin gave higher responses to the TG2-treated avenin than to TG2-treated gluten in four of five patients (CD422, CD431, CD482, and CD496, but not CD507; [Table 2](#pmed-0010001-t002){ref-type="table"}). Notably, intestinal T-cells specific for TG2-treated wheat gluten and gliadin were identified in control biopsies challenged with gluten in all nine celiac patients (see [Table 1](#pmed-0010001-t001){ref-type="table"}). ::: {#pmed-0010001-t002 .table-wrap} Table 2 ::: {.caption} ###### T-Cell Responses to Avenin, Gluten, and Avenin Peptides in T-Cell Lines Established from Biopsies Stimulated with Avenin Antigen ::: ![](pmed.0010001.t002) Both untreated and TG2-treated antigens were tested for recognition. The protein antigens were tested at 100 μg/ml, and the peptide antigens were tested at 10 μM. The responses are given as SIs (\[T + APC + antigen\] divided by \[T + APC\]), and positive responses are indicated by bold type ::: Identification of a T-Cell Epitope in Avenin {#s3c} -------------------------------------------- To identify the T-cell stimulatory peptides, we initially studied an avenin-specific T-cell line (TCL CD422.2.4) isolated from the oat-intolerant patient CD422. The T-cells weakly recognized one gel filtration fraction (\#25) of a pepsin-trypsin digest of avenin. This fraction was further separated by reverse-phase HPLC, and retested for T-cell recognition ([Figure 2](#pmed-0010001-g002){ref-type="fig"}A and [2](#pmed-0010001-g002){ref-type="fig"}B). Stimulatory fractions were subjected to electrospray ionization tandem mass spectrometry ([Figure 2](#pmed-0010001-g002){ref-type="fig"}C). For fractions 3 and 4, a single 22-mer peptide was identified differing only by an asparagine (fraction 3) and an aspartic acid residue (fraction 4). The two identified peptides from fraction 8 represent C-terminally elongated derivates of these 22-mers. Five peptides identified from fraction 9 correspond to N-terminally truncated and C-terminally elongated variants. ::: {#pmed-0010001-g002 .fig} Figure 2 ::: {.caption} ###### Identification of an Epitope in Avenin Recognized by Intestinal T-Cells of Celiac Disease Patients \(A) Reverse-phase HPLC of a pepsin-trypsin digest of avenin. Peptides in fraction 25, obtained from gel filtration, were eluted by an acetonitrile gradient (straight line), and 41 fractions were collected. Fractions recognized by T-cell lines in subsequent experiments are indicated by the numbers above the peaks. \(B) T-cell recognition of fractions obtained by reverse-phase HPLC. All 41 fractions obtained in (A) were tested for recognition by the intestinal T-cell line 422.2.4 (derived from an oat-intolerant celiac disease patient). The fractions were incubated with DR3-DQ2 homozygous antigen-presenting cells overnight before the T-cell line was added. Specific T-cell responses were measured by ^3^H-thymidine incorporation. Pepsin-trypsin-digested avenins, both TG2-treated and untreated, were used as control antigens. \(C) Sequences of the peptides in the stimulatory fractions from reverse-phase HPLC identified by tandem mass spectrometry and overlapping synthetic peptides used for T-cell assays. For better comparison, the amino acid sequence of the avenin precursor protein JQ1047 (gi82331) is taken as a consensus sequence, and deviating residues are underlined. ::: ![](pmed.0010001.g002) ::: T-Cell Recognition of Synthetic Avenin Peptides {#s3d} ----------------------------------------------- Four avenin peptides (1488, 1489, 1490, and 1491; [Figure 2](#pmed-0010001-g002){ref-type="fig"}C) almost completely covering the sequences identified in the reverse-phase HPLC fractions 3, 4, 8, and 9 were synthesized and tested for T-cell recognition. Only peptide 1490 (SEQYQPYPEQEPFVQQQQ) was recognized by the T-cell lines CD422.2.4, CD496.2.1, and CD431.2 (see [Table 2](#pmed-0010001-t002){ref-type="table"}; [Figure 3](#pmed-0010001-g003){ref-type="fig"}). The recognition of this peptide by the T-cell lines CD422.2.4 and CD431.2 was dependent on TG2 treatment. T-cell line CD496.2.1 responded to the native peptide, but the response was augmented by treatment with TG2. We identified one deamidation site by tandem mass spectrometry (underlined in the above given sequence). This regioselectivity of deamidation conforms to the previously defined specificity of TG2 \[[@pmed-0010001-b11],[@pmed-0010001-b19]\]. Several truncation variants of peptide 1490 were also synthesized. The shortest peptides tested were 12-mers; the T-cell line CD431.2 recognized peptide 1505 (YQPYPEQQEPFV) after TG2 treatment and the already deamidated peptide 1504 (YQPYPEQEEPFV) without TG2 treatment ([Figure 3](#pmed-0010001-g003){ref-type="fig"}). We predict the 9-mer core region binding to DQ2 as PYPEQEEPF, placing the glutamic acid resulting from the deamidation at the P6 position. This is similar to the DQ2-α-I gliadin epitope (PFPQPELPY), which also binds to DQ2 with a glutamic acid at the P6 position \[[@pmed-0010001-b20],[@pmed-0010001-b21]\]. Recently, Vader et al. studied whether T-cells generated from celiac disease biopsies stimulated with wheat gluten would cross-react with predicted epitopes of barley, rye, and oats. From these studies they found two broadly reactive polyclonal T-cell lines that responded to peptides from barley hordeins, rye secalins, and the avenin-derived peptides Av-α9A, which is identical to a length variant of 1490 (QYQPYPEQQEPFVQ), and Av-α9B (QYQPYPEQQQPFVQ) \[[@pmed-0010001-b22]\]. We tested peptide Av-α9B against our T-cell lines and found that it was recognized by T-cell lines from the patients CD422 (line 2.4), CD496 (lines 2.1 and 2.3), and CD507 (line 2.3) after TG2 treatment ([Table 2](#pmed-0010001-t002){ref-type="table"}; unpublished data). From the T-cell line CD496.2.3, we generated a T-cell clone that was specific for the peptide Av-α9B after TG2 treatment ([Figure 4](#pmed-0010001-g004){ref-type="fig"}A). This clone responded also to TG2-treated avenin, but did not display cross-reactivity to TG2-treated gluten nor to the TG2-treated 1490 peptide ([Figure 4](#pmed-0010001-g004){ref-type="fig"}B). The avenin-reactive T-cell line generated from the patient CD482 (CD482.2.1) did not recognize the 1490 peptide nor the Av-α9B peptide. Thus, there exist at least two distinct peptides of oats that can elicit mucosal T-cell responses in celiac disease patients with clinical intolerance to oats. ::: {#pmed-0010001-g003 .fig} Figure 3 ::: {.caption} ###### HLA Restriction and Avenin Peptide Specificity of the Intestinal T-Cell Line 431.2 from Patient CD431 The avenin antigen (at 0.25 mg/ml) and peptides (at 10 μM), treated with TG2 when indicated, were incubated overnight with DR3-DQ2 homozygous antigen-presenting cells before T-cells were added. In the HLA restriction experiments, anti-DR or anti-DQ monoclonal antibodies were added 30 min prior to the T-cells. T-cell responses were measured by ^3^H-thymidine incorporation and are represented as SIs. ::: ![](pmed.0010001.g003) ::: ::: {#pmed-0010001-g004 .fig} Figure 4 ::: {.caption} ###### Reactivity of an HLA-DQ2-Restricted T-Cell Clone Derived from a T-Cell Line (CD496.2.3) Established by Avenin Stimulation of an Intestinal Biopsy of Patient CD496 T-cell responses were measured by ^3^H-thymidine incorporation and are represented as SIs. \(A) The clone specifically recognizes the avenin peptide Av-α9B after treatment with TG2. The peptides were tested at 10 μM. \(B) The clone recognizes avenin but not gluten antigen after treatment with TG2. ::: ![](pmed.0010001.g004) ::: Location of Epitopes to a Proline- and Glutamine-Rich Region of Avenin {#s3e} ---------------------------------------------------------------------- The avenin epitopes we identified are localized to a region of avenin uniquely rich in proline and glutamine residues ([Figure 5](#pmed-0010001-g005){ref-type="fig"}). The presumed 9-mer core region of the avenin epitopes (PYPEQQEPF and PYPEQQQPF) contains three proline residues. The high number of proline residues and the localization of the epitopes to a region rich in proline and glutamine residues bear strong resemblance to features typical of DQ2-restricted T-cell epitopes of wheat gliadin \[[@pmed-0010001-b23],[@pmed-0010001-b24]\]. ::: {#pmed-0010001-g005 .fig} Figure 5 ::: {.caption} ###### Amino Acid Sequence of an Avenin (gi 82331, JQ1047) and an α-Gliadin (α2-Gliadin, AJ133612) The proline and glutamine residues are red and blue, respectively. In the avenin, the presumed 9-mer core region of the characterized T-cell epitope is underlined. In the α-gliadin, a 33-mer natural fragment containing six copies of three partly overlapping epitopes (DQ2-α-I, PFPQPQLPY; DQ2-α-II, PQPQLPYPQ; and DQ2-α-III, PYPQPQLPY) is underlined. Note the localization of all the epitopes to regions of the proteins rich in proline and glutamine residues and the high number of proline residues within the 9-mer core regions of the epitopes. ::: ![](pmed.0010001.g005) ::: Discussion {#s4} ========== A number of previous reports concluded that all celiac disease patients tolerate oats. These reports have formed the basis for approving oats in the gluten-free diet for the treatment of celiac disease. The findings reported here demonstrate that oat intolerance exists in some celiac disease patients, and the study provides a molecular explanation for this intolerance. Oats are less related to wheat than are barley and rye. In oats, the prolamines represent much less of the total seed proteins than in the other cereals \[[@pmed-0010001-b25]\]. In addition, avenins contain about half the amount of proline residues (10%) as the prolamins of wheat (gliadins and glutenins), barley (hordeins), and rye (secalines). On this basis, it is intriguing that the identified avenin epitopes are located in the regions of avenins with the highest content of proline residues, regions also rich in glutamine. This is analogous to the localization of the T-cell epitopes in α- and γ-gliadins \[[@pmed-0010001-b23]\]. The immunogenicity of gliadin peptides is influenced both by the glutamine residues, which become specifically deamidated by TG2, and by the proline residues, which protect the peptides from proteolysis in the gastrointestinal tract, determine the specificity of TG2, and are crucial for the selective binding to HLA-DQ2 \[[@pmed-0010001-b01]\]. This study shows that the same features apply to T-cell epitopes of avenin. In humans it is impossible to directly demonstrate that T-cells induce disease. In celiac disease this relates equally to T-cells reactive to gluten and to T-cells reactive to avenin. The fact that avenin-reactive intestinal T-cells, like gluten-reactive T-cells from celiac disease patients, are uniquely restricted by HLA-DQ2 and are activated by TG2-treated peptides speaks strongly in favor of their involvement in the disease pathogenesis. The finding of avenin-specific intestinal T-cells also in individuals with celiac disease that are clinically tolerant to oats does not, as we see it, contradict this assumption. Some patients with celiac disease stay in remission for extended time periods during gluten challenge even if it is likely that they have gluten-reactive T-cells in their intestinal mucosa. Since avenin is less immunogenic than wheat gluten, one would expect an extended time for relapse to be at least as common during oats consumption. It is highly unlikely that the intolerance and the mucosal inflammation observed in our patients could be explained by contamination of the oat flour by wheat, barley, or rye proteins. All the oats consumed were produced in a quality-assessed production line. Our data indicate that avenin can drive mucosal inflammation in that the incubation of the intestinal biopsies with avenin enriches for activated, avenin-reactive T-cells. A substantial proportion of the avenin-reactive T-cells appear to be specific to avenin. The T-cell clone we established from an avenin-challenged biopsy was reactive to avenin but did not cross-react to wheat gluten, and the T-cell lines from biopsies challenged with avenin responded more strongly to avenin than to gluten in four of five participants. Cross-reactivity at the T-cell clonal level has been demonstrated between wheat gluten, hordein, and secalin antigens \[[@pmed-0010001-b22],[@pmed-0010001-b26]\] and likely also exists between gluten and avenin \[[@pmed-0010001-b22]\]. Even if some of the avenin-reactive T-cells were originally primed to gluten and responded to avenin because of cross-reactivity, they would still participate in an avenin-driven immune response. T-cell reponses to the avenin epitopes described in this paper have been found in T-cell lines derived from intestinal biopsies of patients with celiac disease that were stimulated with gliadin \[[@pmed-0010001-b22]\]. It is unknown whether any of the patients from whom these T-cells were isolated had clinical symptoms or mucosal inflammation related to oats ingestion. Thus, to our knowledge, the current study is the first to demonstrate a mechanistic link between clinical symptoms of oat intolerance, mucosal inflammation, and avenin-reactive T-cells. Oat intolerance can cause complications in the large group of celiac disease patients who are now regularly consuming oats. At this stage we do not know how frequently such complications may occur. Presumably such complications will not be very common, but only extended clinical follow-up of oats-consuming celiac disease patients will establish the frequency. Monitoring of T-cell responses to avenin epitopes may potentially identify individuals who are at risk of developing oat intolerance. Based on our data, such monitoring will also identify some individuals who are clinically tolerant to oats and who have minimal or no mucosal pathology after a limited oats challenge. Possibly some of these patients may have latent oat intolerance that will develop into overt disease after prolonged exposure, but this remains speculative. Our observations demonstrate that even if oats seem to be well tolerated by many celiac disease patients, there are patients who have an intestinal T-cell response to oats. Until the prevalence of oat intolerance in celiac disease patients is established, clinical follow-up of celiac disease patients eating oats is advisable. Clinicians should be aware that oat intolerance may be a reason for villous atrophy and inflammation in patients with celiac disease who are eating oats but otherwise are adhering to a strict gluten-free diet. ::: {.caption} ###### Patient Summary ::: {#sb1a} ### Background {#sb1a1} Celiac disease is a digestive disease that damages part of the gut (the small intestine) and interferes with absorption of nutrients from food. Patients with celiac disease do not tolerate a protein called gluten, which is found in wheat, rye, and barley. When people with celiac disease eat foods containing gluten, their immune system responds by damaging the small intestine. The disease is quite serious in some patients, but eating a strictly gluten-free diet can eliminate all of the symptoms. Unfortunately, wheat, barley, and rye products like flour are found in many common foods, and patients have to avoid them for the rest of their lives. Previous studies suggested that oats were safe for patients with celiac disease, and as a result, they often form part of a gluten-free diet. ### What Did the Researchers Find? {#sb1a2} Contrary to other studies, this one demonstrates that oats intolerance does exist in some patients with celiac disease. These patients have an immune reaction to oats that is similar to the reaction most celiac disease patients have to wheat, barley, and rye. ### What Does This Mean for Patients? {#sb1a3} It appears that oats are not safe for all patients with celiac disease. Patients who eat oats as part of a gluten-free diet should discuss their diet and any symptoms with their doctors; doctors should keep in mind that patients might develop symptoms when they eat oats. ### What Are the Problems with the Study? {#sb1a4} The researchers studied only a small number of patients, and this study cannot tell us how common oats intolerance is among celiac disease patients. ### Where Can I Find More Information? {#sb1a5} US National Institutes of Diabetes, Digestive, and Kidney Disorders: <http://digestive.niddk.nih.gov/ddiseases/pubs/celiac/> Celiac Disease Foundation: <http://www.celiac.org/> The Gluten Intolerance Group: <http://www.gluten.net/> The Celiac Disease Foundation: <http://www.celiac.com/> This work is supported by the Research Council of Norway, the European Commission (project QLK1-2000-00657), the Norwegian Foundation for Health and Rehabilitation, the Deutsche Forschungsgemeinschaft (SFB 510, Project D4), ZonMW (grant 912-02-028), and the Stimuleringsfonds Voedingsonderzoek LUMC. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We thank Jon Matre and Even Lind for referring two of the oat-intolerant patients to Rikshospitalet University Hospital; Marie Kongshaug Johannesen, Elisabeth Reed Eng, Eva Boretti, and Nicole Sessler for excellent technical assistance; and Don Kasarda and Chaitan Khosla for comments on the manuscript. Citation: Arentz-Hansen H, Fleckenstein B, Molberg Ø, Scott H, Koning F, et al. (2004) The molecular basis for oat intolerance in celiac disease patients. PLoS Med 1(1): e1. IEL : intraepithelial lymphocyte SI : stimulatory index TG2 : tissue transglutaminase [^1]: Competing Interests: The authors have declared that no competing interests exist. LMS is a member of the editorial board of PLoS Medicine. [^2]: Author Contributions: ØM, KEAL, and LMS designed the study. HAH, BF, ØM, HS, FK, GJ, PR, KEAL, and LMS analyzed the data. FK and GJ contributed synthetic peptides for the study. HAH, BF, ØM, HS, FK, GJ, KEAL, and LMS contributed to writing the paper.	PubMed Central	2024-06-05T03:55:48.571444	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523824/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e1", "authors": [ { "first": "Helene", "last": "Arentz-Hansen" }, { "first": "Burkhard", "last": "Fleckenstein" }, { "first": "Øyvind", "last": "Molberg" }, { "first": "Helge", "last": "Scott" }, { "first": "Frits", "last": "Koning" }, { "first": "Günther", "last": "Jung" }, { "first": "Peter", "last": "Roepstorff" }, { "first": "Knut E. A", "last": "Lundin" }, { "first": "Ludvig M", "last": "Sollid" } ] }
PMC523825	The obstetric urogenital fistula has caused women misery ever since they first started delivering children. It was once common worldwide, but with the advent of safe obstetric care during the early part of the last century, the condition has become rare in rich countries. Urogenital fistulae do still occur in developed countries, but unlike in the developing world, they are usually a complication of a difficult pelvic surgery, cancer, or radiation \[[@pmed-0010002-b2]\]. Obstetric Fistula: A Disease of Poverty {#s2} ======================================= Obstetric fistula---a urogenital fistula from obstructed labour---is now only encountered in countries where health resources are scarce. The shame associated with incontinence drives affected women further into a life of poverty and begging. Many women with fistula either do not know that they can get medical help, or if they do, they are unable to pay. Furthermore, very little scientific research has been published about obstetric fistula and its management, partly because the people treating patients with this condition are working in remote areas, often with very limited resources for research. What has been written consists largely of personal case series and a few epidemiological studies \[[@pmed-0010002-b2]\]. To date there has been only one randomised trial in the developing world, involving 79 women operated on by a single surgeon in Benin, which found that intra-operative intravenous antibiotics did not reduce the risk of failed surgical repair or of objective incontinence \[[@pmed-0010002-b3]\]. There has been only one study in a developing country comparing different surgical techniques---a retrospective study of 46 patients operated on over a five-year period at a hospital in Mumbai, India \[[@pmed-0010002-b4]\]. This study suggested that a technique called the Martius procedure (which involves grafting of a labial pad of fat) may be better than simple anatomic repair. What we do know about the obstetric urogenital fistula is that the women who have these injuries are young, usually illiterate, and of a lower socioeconomic background. They are more often primiparous and short in stature, and they have an average length of labour of some 3.9 days. The labour is usually unattended, or if attended, it is by someone unskilled. The women inevitably deliver a stillborn child. About half of the women with fistula are divorced as a direct result of their incontinence \[[@pmed-0010002-b2],[@pmed-0010002-b5],[@pmed-0010002-b6],[@pmed-0010002-b7]\]. The Injury and its Consequences {#s3} =============================== The initial injury that leads to a fistula results from ischaemic necrosis of the soft tissues of the pelvis due to an impacted presenting part during the long labour.[](#pmed-0010002-g001){ref-type="fig"} ::: {#pmed-0010002-g001 .fig} ::: {.caption} ###### Incontinent women face a life of shame and isolation (Photograph: © 2004 Shaleece Haas. This is an open-access image distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.) ::: ![](pmed.0010002.g001) ::: The ischaemia then affects the bladder and vagina (and sometimes the rectum and vagina), resulting in a fistula. The process also affects other pelvic structures. These include the nerves of the sacral plexus, resulting in foot drop and hamstring compartment weakness (foot drop may also be a result of prolonged squatting in labour, injuring the common peroneal nerve as it traverses the head of the fibula). Bony abnormalities are common, separating or obliterating the symphysis pubis. Up to half of patients develop upper renal tract abnormalities: scarring of the ureter can cause obstructive uropathies \[[@pmed-0010002-b8]\]. Up to two thirds of women are rendered amenorrhoeic (their periods stop), either from disorders of the hypothalamic-pituitary axis or from Asherman syndrome (adhesions in the uterus due to scarring). The vagina may be completely destroyed, as may the cervix, causing an obstructive outflow tract resulting in cryptomenorrhoea (women menstruate, but the sloughed blood and tissue don\'t leave the body). The continual leakage of urine over the perineal skin can cause local and painful irritation, termed 'urine dermatitis'. Bladder stones can occur, as women affected by fistula often drink less to try and pass less urine and the concentrated urine can form calculi. The obstetric injury has been termed a 'field injury', as the pathology is broad rather than isolated \[[@pmed-0010002-b9]\]. The resulting range of injuries can be daunting for health professionals who are working with limited resources. The Ilorin Experience {#s4} ===================== A recently published retrospective case note review provides new data on obstetric urogenital fistula in northern Nigeria. Ijaiya and Aboyeji reviewed 34 cases of fistula managed at the University of Ilorin Teaching Hospital over a two-year period \[[@pmed-0010002-b1]\]. During this period, there were 32,188 deliveries---thus, the incidence of fistula was 1.1 per 1000 live births. The mean age of the women with fistula was 23.9 years, and 32 of the 34 women were illiterate. Half were primiparous. The most common cause of the fistula was obstructed, prolonged labour---the cause in 28 out of the 34 cases. The most common complications of the fistulae were divorce or separation (eight women) and amenorrhoea (seven women). ::: {.caption} ###### What Is Obstetric Fistula? ::: '\[Obstetric fistula\] usually occurs when a young, poor woman has an obstructed labour and cannot get a Caesarean section when needed. The obstruction can occur because the woman\'s pelvis is too small, the baby\'s head is too big, or the baby is badly positioned. The woman can be in labour for five days or more without medical help. The baby usually dies. If the mother survives, she is left with extensive tissue damage to her birth canal that renders her incontinent.' Source: UNFPA Campaign to End Fistula: "What is Fistula?" ([www.unfpa.org/fistula/about.htm](www.unfpa.org/fistula/about.htm)). How does this study compare with other literature on obstetric fistula in Nigeria? First, the incidence reported in the study is lower than that of another hospital study of 22,774 deliveries in Zaria, also in northern Nigeria, which gave an incidence of 3.5 per 1000 deliveries \[[@pmed-0010002-b10]\]. However, both of these incidence figures are from hospital-based studies, and it is thought that most women do not get to a health facility to deliver their child. So the true incidence of obstetric fistula may well be much higher. In terms of prevalence, it has been estimated that there are up to 800,000 women in Nigeria who have a urogenital fistula from obstructed labour \[[@pmed-0010002-b11]\]. Second, although the figures given in Ijaiya and Aboyeji\'s study differ slightly from those of other publications, their study does reconfirm the trends in aetiology and epidemiology of obstetric fistula in the developing world.[](#pmed-0010002-g002){ref-type="fig"} ::: {#pmed-0010002-g002 .fig} ::: {.caption} ###### Surgery at the Fistula Hospital, Addis Ababa (Photograph: © 2004 Shaleece Haas. This is an open-access image distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.) ::: ![](pmed.0010002.g002) ::: Treatment and Prevention {#s5} ======================== As in other resource-poor countries, many women with obstetric fistula in Nigeria do not get to a surgeon with expertise in fistula repair. There are, however, a few dedicated professionals in Nigeria helping women with fistula and operating on up to 1,600 women a year \[[@pmed-0010002-b11]\]. Resource-rich countries were able to eradicate the obstetric fistula almost 100 years ago, but the challenge to resource-poor countries is enormous. There are an estimated 2 million women with fistula in the world, with anywhere between 100,000 and 500,000 new cases developing each year \[[@pmed-0010002-b11]\]. At the world\'s current capacity for dealing with the problem, it would take up to 400 years to treat the backlog of patients. Clearly we need many more centres equipped to care for women with fistula. The United Nations Population Fund (UNFPA; [www.unfpa.org](www.unfpa.org)) and the International Federation of Gynecology and Obstetrics ([www.figo.com](www.figo.com)) are endeavouring to help. UNFPA has already sponsored training workshops on fistula surgery for surgeons and fledgling fistula units in Bangladesh and some parts of Africa. The obstetric fistula is an entirely preventable condition. Several strategies have been proposed to eradicate this condition in developing countries ([Box 1](#box1){ref-type="boxed-text"}), just as it has been eradicated in the developed world. However, to prevent any new cases of obstetric fistula from occurring, there would need to be 75,000 new emergency obstetric centres built in Africa alone \[[@pmed-0010002-b12]\]. This would require not only funds, but an appropriate number of trained doctors, nurses, midwives, and support personnel. ::: {.caption} ###### Box 1. The UNFPA\'s Key Strategies to Address Fistula ::: 'Postpone marriage and pregnancy for young girls'Increase access to education and family planning services for women and men'Provide access to adequate medical care for all pregnant women and emergency obstetric care for all who develop complications'Repair physical damage through medical intervention and emotional damage through counselling\'Source: UNFPA Campaign to End Fistula: "Fast Facts" ([www.unfpa.org/fistula/facts.htm](www.unfpa.org/fistula/facts.htm)). Even if such centres are established, women will need to be convinced of the importance of seeking help without delay for a difficult labour. And then, to be able to receive that help, roads need to be built, transport systems need to be put in place, and communications need to be improved. The obstacles are clearly huge, and with currently very little money and very few professionals available, women with obstetric fistula will sadly be with us for many more years to come. Supporting Information {#s6} ====================== Text S1 ::: {.caption} ###### Full Text of Ijaiya and Aboyeji\'s Study \[[@pmed-0010002-b1]\] (234 KB PDF). ::: ::: {.caption} ###### Click here for additional data file. ::: Citation: Browning A (2004) Obstetric fistula in Ilorin, Nigeria. PLoS Med 1(1): e2. UNFPA : United Nations Population Fund [^1]: Andrew Browning is a Staff Specialist at the Addis Ababa Fistula Hospital, Addis Ababa, Ethiopia. E-mail: <fistula-hospital@telecom.net.et> [^2]: Competing Interests: The author declares that he has no competing interests.	PubMed Central	2024-06-05T03:55:48.573906	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523825/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e2", "authors": [ { "first": "Andrew", "last": "Browning" } ] }
PMC523826	At the World Health Assembly in May 2003, three civil society groups---the People\'s Health Movement, the Global Equity Gauge Alliance, and Medact---discussed the need for civil society to produce its own alternative to the World Health Organisation\'s World Health Report. We felt strongly that we needed to produce a global health report that had equity and the right to health at its heart. We also needed a way to monitor the performance of global health institutions themselves. The idea of an alternative to the World Health Report has developed into an initiative called the Global Health Watch, which we are launching next year. The Three Key Players {#s2} ===================== Medact (<http://www.medact.org>) is a United Kingdom--based global health charity, undertaking education, research, and advocacy on conflict, poverty, and the environment. The Global Equity Gauge Alliance (<http://www.gega.org.za>) was created to participate in and support an active approach to monitoring health inequalities and promoting equity within and between societies. The Alliance currently includes 11 member-teams, called Equity Gauges, located in ten countries in the Americas, Africa, and Asia. The People\'s Health Movement (<http://www.phmovement.org>) is a global network of activists, organisations, and social movements. Its goal is to re-establish health and equitable development as top priorities in local, national, and international policy-making, with comprehensive primary health care as the strategy to achieve these priorities. Why an Alternative Is Needed {#s3} ============================ Concerted action by civil society has had tremendous success in the field of international health---global grassroots campaigns on infant feeding, smoking, and drug prices have changed policies and people\'s lives.[](#pmed-0010003-g001){ref-type="fig"} ::: {#pmed-0010003-g001 .fig} ::: {.caption} ###### The Watch: Examining the world\'s health from an alternative perspective (Illustration: Giovanni Maki, Public Library of Science) ::: ![](pmed.0010003.g001) ::: But over the last two decades---at the same time as these campaigns have scored victories---there has, in some parts of the world, been a stagnation and even reversal of the dramatic gains in life expectancy witnessed by many others for much of the 20th century. These reversals, unprecedented outside times of war and famine since the early 1800s and a scandal in a world of enormous wealth and technological prowess, have once more thrown the spotlight on how underlying social and economic problems affect health and health services. The setbacks have also underlined appalling failures of health development policy. Ambitious targets to achieve 'Health for All' agreed to at the end of the 1970s by health ministers from around the world have failed miserably; a similar fate seems likely for the targets set out in the Millennium Development Goals for 2015. As a result, there are large question marks hanging over the effectiveness of international health policy. These are the reasons why we have decided to produce Global Health Watch, which we hope will become a regular report on international health issues ([Box 1](#box1){ref-type="boxed-text"}). We believe that civil society campaigners need to look at the broader health agenda---beyond single-issue advocacy. Major concerns about health systems such as poor pay and working conditions for health professionals, creeping commercialisation, and plummeting public investment have not had the attention they deserve. Likewise, broader determinants of health---such as education, water, food, and the environment---are often insufficiently regarded when health policies are formulated. The Watch attempts to focus minds on the need for more integrated planning across sectors and on the creation of health systems that promote social justice rather undermine it. ::: {.caption} ###### Box 1. Global Health Watch---2005 Report Contents ::: Section A: The Politics and Economics of Health in the 21st Century Section B: The Health Care Sector Health systems that promote social justiceResponding to the commercialisation of health careThe pharmaceutical industry, access to medicines, and intellectual property rightsHuman resources: the lifeblood of health systemsResponding to HIV/AIDSGene technology and the attainment of health for allSexual and reproductive health Section C: Beyond Health Care Environmental challengesMilitarism and conflictWaterThe right to food: land, agriculture, and household food securityEducation Section D: Marginalised Groups Indigenous peoplesPeople with disabilities Section E: Monitoring of Institutions and Resource Flows World Health OrganisationWorld BankWorld Trade Organisation and trade agreementsGlobal Fund and Pepfar (United States President\'s Emergency Plan for AIDS Relief)Monitoring of international promises on aid and debt relief Section F: Summary and Strategies for Action How Will the Watch Be Different? {#s4} ================================ This is how the Watch will be alternative: it will present options for health policy-makers that question the dominant reform agenda that emphasises market-driven and diseased-based approaches to health care. A policy bias against government action and a lack of creative thinking about how governments can shape health care markets to work in favour of equity and social inclusion are unfortunate features of global health debates. More recently, the emphasis has been placed once again on campaigns against specific diseases such as HIV/AIDS and tuberculosis, despite the universally acknowledged importance of building and maintaining health systems that can respond to the broader needs of patients. We hope the Watch will present some alternative and imaginative thinking about how health services can respond creatively to the many challenges they face, with a strong focus on basic principles of equity and universality and avoiding top-down disease-focussed programmes that neglect the broader determinants of health. We have invited some of the most interesting and innovative thinkers in health policy---from both developing and developed countries and from academia and civil society---to help us achieve these objectives. The Watch will also be 'alternative' in another sense---it will act as a regular monitor of the policies, governance, and funding of the institutions affecting global health, including the World Health Organisation and World Bank, something no other health report undertakes. We hope to offer proposals for reform, as well as to stimulate further action by civil society to make these institutions more accountable and responsive to the needs of the poor and vulnerable. Linking Civil Society Groups {#s5} ============================ It is important to say that the three networks and organisations that have convened the Watch are really just its initiators. In the end we hope the Watch will be backed by as many individuals, organisations, and social movements as possible, strengthening the links between civil society organisations across countries and across health-related sectors, and increasing the power and influence of the report itself. Already, many have expressed their interest in the project, and their willingness to contribute: through writing chapters, contributing case studies, and launching the Watch and promoting it in their country when it is finished. Groups from India and Brazil are planning parallel national Watches. We plan to launch the Watch at the second People\'s Health Assembly, which will be held in Ecuador in July 2005. We don\'t want this report to be addressed just to health activists or health policy-makers or academics. If we are going to create change we need to capture the imagination of the broader health professional community and the public at large. That is why we encourage readers to get involved and tell others about the Watch and to use it to throw down a challenge to those who call the shots at national and international levels. Citation: Rowson M, McCoy D, Gupta AS, de Negri Filho A (2004) The Global Health Watch. PLoS Med 1(1): e3. [^1]: Mike Rowson is the executive director of Medact, United Kingdom. David McCoy works for the Secretariat of Global Equity Gauge Alliance, South Africa. Amit Sen Gupta is the co-convener of the People\'s Health Movement, India. Armando de Negri Filho is the general coordinator of the Latin American Association of Social Medicine, Brazil, and is president of the International Society for Equity in Health. [^2]: Competing Interests: The authors declare that they have no competing interests.	PubMed Central	2024-06-05T03:55:48.575070	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523826/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e3", "authors": [ { "first": "Mike", "last": "Rowson" }, { "first": "David", "last": "McCoy" }, { "first": "Amit Sen", "last": "Gupta" }, { "first": "Armando", "last": "de Negri Filho" } ] }
PMC523827	Ann Taket\'s Viewpoint: Routinely Asking about Domestic Violence Is Worthwhile {#s1} ============================================================================== Domestic violence is a misunderstood topic. The context of a trusted health professional talking to a woman is one that provides an important opportunity for providing information to counter misconceptions. I deliberately talk about this in terms of asking all women about domestic violence and not in terms of screening women for domestic violence. It is not appropriate or helpful to regard enquiry about being abused as a form of screening. Domestic violence is not a disease present in the body of the person who experiences it---rather it is a health-related risk factor. As such, knowledge of abuse puts health professionals in a position to respond better to the needs of women affected by it. Professionals can respond by providing information on specialist services---usually provided outside the health service---that women may access if they wish. By giving information to affected women, health professionals can also help to reduce women\'s sense of isolation and stigmatisation. Asking about experience of domestic violence can be seen as a routine part of history taking, just as health professionals regularly and repeatedly ask patients about their smoking behaviour, alcohol use, weight, and exercise. The prevalence of domestic violence among women is such that, even if it is not a personal issue for the woman concerned, it most likely will be for one or more of her relatives, friends, and neighbours \[[@pmed-0010004-b1]\]. Since many women experiencing abuse feel alone and ashamed, and their abusers often encourage them to believe that the abuse is their fault, presenting information to counter women\'s negative feelings is an important preventive strategy. Most women experiencing domestic violence report that the specialised services that exist to respond to their needs were difficult to find out about \[[@pmed-0010004-b2]\]. The provision of simple information on the existence of specialised services and how to contact them is relevant to all women. Studies have examined women\'s views on being asked about domestic violence. These studies have shown that once they have experienced being asked, they are usually in favour of being asked. This is true both for those who have experienced or are experiencing abuse, and those who have not \[[@pmed-0010004-b3]\]. It is only a small minority of women who object to being asked, or who find the question uncomfortable. Women who have experienced abuse particularly value being asked directly.[](#pmed-0010004-g001){ref-type="fig"} ::: {#pmed-0010004-g001 .fig} ::: {.caption} ###### Women experiencing violence often feel alone and ashamed (Illustration: Margaret Shear, Public Library of Science) ::: ![](pmed.0010004.g001) ::: Asking about abuse should be done in a flexible fashion---the particular questions used should respond to the circumstances of the consultation. For example, it is appropriate to ask women about domestic violence as part of a health check in a Well Woman Clinic, but it would be completely inappropriate in a consultation where another adult or a child was present. By being flexible, health professionals can integrate their questioning within a variety of different encounters. Integrating questions about abuse into routine encounters provides for the maintenance of confidentiality and safety. In order to do this, health professionals require training on raising the issue and knowledge about local advice and support services. Committees on both sides of the Atlantic have rejected the notion of screening women for domestic violence, arguing that there is insufficient evidence of the effectiveness of interventions \[[@pmed-0010004-b4],[@pmed-0010004-b5]\]. Part of the reason for this lack of evidence is that the systematic reviews on which these committees based their recommendations often excluded the most important types of evidence that do exist \[[@pmed-0010004-b3],[@pmed-0010004-b6]\]. For example, these reviews excluded studies done outside the health service setting---they excluded those based in social services, or in the voluntary or community sector. Some excluded studies show the effectiveness of specialised service provision for women experiencing abuse. In one example of an excluded study, researchers used a randomised design to evaluate an advocacy service for women experiencing domestic violence \[[@pmed-0010004-b7],[@pmed-0010004-b8]\]. Women were interviewed six times over two years, and women in the intervention group reported a higher quality of life, decreased difficulty in obtaining community resources, and less violence over time than women in the control group. Other studies showing the value of specialised support services provided outside of the health system provide evidence of the potential benefits of asking about abuse \[[@pmed-0010004-b2]\]. Systematic reviews have also excluded, or devalued, evidence from qualitative studies. For example, a study of 200 women who had used domestic violence outreach services found that about half were living in situations of domestic violence when they first contacted the service. All of these women reported that the outreach services had helped them to leave the abusive relationship---a valued outcome for them \[[@pmed-0010004-b9]\]. Given the health impacts on women who experience domestic violence (not to mention their children) and the prevalence of the problem, routinely asking women about abuse should be seen as an important form of primary and secondary prevention for a wide range of health problems. Nadine Wathen and Harriet MacMillan\'s Viewpoint: The Decision to Screen Should Be Based on Evidence {#s2} ==================================================================================================== Screening tools for domestic violence are abundant, and many are effective at identifying women experiencing abuse \[[@pmed-0010004-b3],[@pmed-0010004-b10]\]. However, merely identifying a woman as abused has not been shown to actually improve her quality of life or reduce the violence she is experiencing \[[@pmed-0010004-b6],[@pmed-0010004-b11]\]. Furthermore, with one exception \[[@pmed-0010004-b7]\], we do not know whether interventions for women exposed to violence are effective in reducing violence or improving other health-related outcomes. Interventions for abusive men have shown little effectiveness \[[@pmed-0010004-b11],[@pmed-0010004-b12]\]. Given the morbidity and mortality associated with domestic violence, it is tempting to suggest that universal screening for abuse should be integrated into routine clinical care, such that all women, regardless of their reason for presenting to a clinical setting, should be "asked the question." Some argue that this approach is justified by the need to increase awareness of domestic violence as a significant problem with serious health and social consequences, and to make abused women aware that they are not alone in their experience. These are important considerations. Certainly all women who disclose that they have been exposed to violence should be provided with options regarding seeking help \[[@pmed-0010004-b13]\]. Good diagnostic assessment requires that clinicians be able to identify and respond to signs and symptoms of abuse, from patterns of physical injury to mental health concerns, including unexplained pain and depression. Not asking women about exposure to violence during certain diagnostic assessments (such as investigation of chronic pain) may lead to misdiagnosis and a path of inappropriate investigations or treatments that will miss the underlying problem \[[@pmed-0010004-b14]\]. It is also imperative that clinicians know about the hospital- or community-based services that exist and ensure that there is a system in place to provide appropriate referral \[[@pmed-0010004-b15]\]. However, what about women presenting without obvious signs and symptoms of domestic violence---such as a woman who comes to the clinic for assessment of an upper respiratory tract infection? Should such women be prompted to disclose whether they are being abused? The woman who is not being abused will answer to that effect, and the appointment can carry on. But for the woman who is experiencing violence, who has not volunteered this information, several factors must be considered. An important issue is whether she is ready---both psychologically and in terms of taking specific actions---to confront the issue. A number of excellent qualitative studies have examined the process that women undertake in acknowledging that they are "victims" of "abuse" and embarking on the often long and difficult journey to avoid, reduce, and ultimately stop the violence in their lives \[[@pmed-0010004-b16],[@pmed-0010004-b17]\]. Given the enormousness of that task, the key question becomes the extent to which prompting disclosures of abuse through universal screening will actually help women in this process, and help them in a way that they find meaningful. Any potential benefits of screening must then be weighed against its potential harms, including labelling women, prompting potentially premature disclosure, and triggering possible reprisal violence from the abuser if he discovers she has sought help. The last of these might be particularly exacerbated for the woman with the respiratory tract infection who was unprepared to disclose and did not take necessary precautions. Other potential harms include exposure to the ramifications of laws on mandatory child protection reporting, whereby health providers must report such disclosures to child protection authorities. This can lead to an investigation that potentially increases a woman\'s risk of exposure to violence, and in some cases of having her children placed in foster care. Research has shown that many of these potential harms are of concern to women when mandatory universal screening and/or reporting protocols are in place \[[@pmed-0010004-b18]\]. Finally, from a health system perspective, the opportunity cost of not having used this time with the woman to conduct screening or prevention activities for which there is proven benefit, such as counselling about Pap smears or mammograms, should not be discounted.[](#pmed-0010004-g002){ref-type="fig"} ::: {#pmed-0010004-g002 .fig} ::: {.caption} ###### There are potential harms from "asking the question" (Illustration: Margaret Shear, Public Library of Science) ::: ![](pmed.0010004.g002) ::: Given the lack of clear data on the benefits of screening and of the interventions to which women are referred, and the lack of data on potential harms, we and others have concluded the following \[[@pmed-0010004-b3],[@pmed-0010004-b19],[@pmed-0010004-b20]\]. Until these questions are answered, the most appropriate health care system approach is the more targeted case-finding or diagnostic method, which focuses health care resources on those in immediate need of care. Our hope is that studies currently underway (for example, those supported by the Ontario Women\'s Health Council and the US Centers for Disease Control) will provide information about the effectiveness of domestic violence screening. Let\'s base the decision about implementation of screening on evaluations of whether such screening does more good than harm in the lives of women. Taket\'s Response to Wathen and MacMillan\'s Viewpoint {#s3} ====================================================== I agree entirely with Nadine Wathen and Harriet MacMillan that practice should be based on evidence. There are further areas of agreement. We agree that there is a lack of knowledge on effective interventions for abusers and on harm occurring as a result of enquiry, and that targeted case finding is important. The key difference that exists between my viewpoint and theirs is the conclusion about whether health professionals should aim to ask all women about domestic violence. Underlying this difference is the issue about how much evidence we need, and of what type. My position is that the evidence that already exists is sufficient to justify the promotion of routine enquiry, aiming to ask all women about their experience of abuse. There is evidence of actual benefits to women---and their children---from interventions provided by specialised services for domestic violence and from brief discussions with health professionals \[[@pmed-0010004-b21]\]. Aiming to ask all women has several advantages over targeted case finding \[[@pmed-0010004-b22]\]. It contributes to changing social attitudes to domestic abuse, it is less likely to make women experiencing abuse feel stigmatised, and it is less likely to compromise the safety of women experiencing abuse. Furthermore, health professionals report that their perceptions about which women are being abused, and which are not, are often wrong. The twin issues of women\'s safety and harm minimisation are extremely important, for both routine enquiry and targeted case finding. These issues are important reasons why training and protocols for enquiry are necessary. Standard principles of confidentiality should be reinforced in training and protocols, which need to be tailored to relevant legal requirements, such as when child protection issues are involved. Training and protocols also need to emphasise that the role of routine enquiry is to facilitate, and not force, disclosure. It must remain the woman\'s choice as to if, when, and to whom, she discloses. Wathen and MacMillan\'s Response to Taket\'s Viewpoint {#s4} ====================================================== We agree with Ann Taket that domestic violence is not a disease, and that the paradigm of "screening for disease" is problematic in this context. At issue, however, is the question of whether domestic violence should be "talked about" with all women or only in situations where asking about it is part of a specific diagnostic assessment. As with screening for a disease, universal screening for domestic violence should not be implemented unless we are sure that interventions are available to help those identified via screening and that screening plus appropriate treatment will do more good than harm. Professor Taket outlines the importance of integrating discussions about abuse in consultations to raise community awareness. Unfortunately, there is no evidence that this type of consciousness-raising occurs, or if it does, what benefit it might have. Given the lack of effectiveness of educational campaigns in general, it is difficult to be optimistic about this approach. We disagree with her conclusion that existing systematic reviews have "excluded studies done outside the health service setting...." Our review included interventions such as the post-shelter advocacy counselling approach to which Professor Taket refers \[[@pmed-0010004-b11]\]. This intervention has been recommended by the Canadian Task Force on Preventive Health Care as one to which, where available, clinicians might refer women in these circumstances \[[@pmed-0010004-b19]\]. However, since shelters themselves have not been adequately evaluated, the value of linking screening to a post-shelter intervention is unclear. Finally, we concur that qualitative studies are invaluable in understanding domestic violence. Such research has provided insight into the complex process that women undertake to address the violence in their lives. Until there is evidence that universal screening actually helps with this process, the focus should be on developing evidence-based approaches to assist women when they do disclose abuse and on training health professionals to respond appropriately to such disclosures. Citation: Taket A, Wathen CN, MacMillan H (2004) Should health professionals screen all women for domestic violence? PLoS Med 1(1): e4. [^1]: Ann Taket is a professor of primary health care at London South Bank University, London, England. E-mail: <a.r.taket@lsbu.ac.uk> [^2]: Nadine Wathen is a research fellow in the Department of Psychiatry and Behavioural Neurosciences at McMaster University, Hamilton, Canada. E-mail: <wathenn@mcmaster.ca> [^3]: Harriet MacMillan is a professor in the Department of Psychiatry and Behavioural Neurosciences and Department of Pediatrics at McMaster University, Hamilton, Canada. E-mail: <macmilnh@mcmaster.ca> [^4]: Competing Interests: Ann Taket declares that she has no competing interests. [^5]: Competing Interests: Nadine Wathen holds a Canadian Institutes of Health Research (CIHR)--Ontario Women\'s Health Council Fellowship. Harriet MacMillan holds research funding from the CIHR Institutes of Gender and Health; Aging; Human Development, Child and Youth Health; Neuroscience, Mental Health, and Addiction; and Population and Public Health, and from the Ontario Women\'s Health Council.	PubMed Central	2024-06-05T03:55:48.576388	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523827/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e4", "authors": [ { "first": "Ann", "last": "Taket" }, { "first": "C. Nadine", "last": "Wathen" }, { "first": "Harriet", "last": "MacMillan" } ] }
PMC523828	"If someone is condemned to a premature death because of the injustice of global health inequality, it is doubly unjust for that person to be condemned to an agonising death racked by preventable pain." \[[@pmed-0010005-b1]\] In many resource-poor countries, death is accompanied by avoidable pain and other distressing symptoms. Unfortunately, governments in these countries usually give care at the end of life a low priority compared with preventive and curative services \[[@pmed-0010005-b2]\]. This prioritization makes little sense, especially when applied to treating patients with cancer and HIV/AIDS, since prevention efforts are often failing to reduce the disease burden, while treatments aimed at cure or prolonging life are still too expensive to be made widely available. As three physicians in Jamaica, Uganda, and Rwanda, we believe that providing quality care at the end of life should be seen as a global public health priority. By using relatively low-cost palliative care approaches and community-based strategies, thousands of terminally ill patients in Africa and the Caribbean could be relieved of their pain and suffering. The Burden of Cancer and HIV/AIDS {#s2} ================================= In the countries where we work, the burden of cancer and HIV/AIDS is overwhelming. In Africa about 2.5 million people die annually from HIV/AIDS, and more than 0.5 million die from cancer \[[@pmed-0010005-b3],[@pmed-0010005-b4]\]. Sepulveda and colleagues have estimated that each year, at least one in 200 people in the five African countries that they studied (Botswana, Ethiopia, Tanzania, Uganda, and Zimbabwe) need palliative care at the terminal stages of HIV/AIDS or cancer \[[@pmed-0010005-b2]\]. This figure does not include those needing palliative care for other diseases or those suffering from a serious illness in the pre-terminal stages. Thus, perhaps one in 100 people in these countries needs some level of palliative care each year \[[@pmed-0010005-b2]\]. In Rwanda, as in most other African countries, infectious diseases are still rife. Health professionals are often faced with the terrible dilemma of having to choose between saving lives and easing the suffering of the dying. Indeed the authorities usually believe that any investment in palliative care would be at the expense of providing life-saving treatments for those suffering from curable, often infectious illness. In many Caribbean countries, while the scourge of water- and insect-borne infectious diseases is largely under control, the prevalence rates of HIV in the adult population are some of the highest in the world \[[@pmed-0010005-b5]\]. In Jamaica, the largest English-speaking country in the Caribbean (population 2.5 million), in 2001, there were an estimated 20,000 people living with HIV and 980 deaths from AIDS \[[@pmed-0010005-b6]\]. Further, Jamaica\'s proximity to the United States means that many people aspire to a lifestyle more representative of a wealthy, industrialized nation, and are thus susceptible to diseases such as cancer, coronary artery disease, and diabetes. Unfortunately, the island\'s struggling public health system is often unable to provide adequately for patients with these diseases.[](#pmed-0010005-g001){ref-type="fig"} ::: {#pmed-0010005-g001 .fig} ::: {.caption} ###### Better distribution of analgesics would improve palliative care provision ::: ![](pmed.0010005.g001) ::: The Arguments for Palliative Care {#s3} ================================= Prevention efforts---including health promotion, education, and screening---and treatments aimed at cure or prolonging life are key strategies needed to reduce the burden of HIV/AIDS and cancer in resource-poor countries \[[@pmed-0010005-b7]\]. However, when it comes to prevention, in many countries the effects of health education, health promotion, and screening programs have yet to make an impact on rates of HIV infection or cancer. When it comes to treatment, the provision of high-quality, affordable treatments for patients with HIV/AIDS and cancer requires the development of appropriate and accessible infrastructure and technology with sustainable funding. At present, access to treatment where we are working is essentially controlled by the ability of the patient to pay. Thus, only about one in 200 people with HIV in Uganda are able to obtain antiretroviral medicines \[[@pmed-0010005-b8]\]. Furthermore, patients in developing countries often present with far advanced malignant disease, and as many as 80% of people with cancer may be incurable at diagnosis \[[@pmed-0010005-b9]\]. Given that prevention isn\'t taking effect in many places, and curative services are poorly available or inappropriate, we believe that the provision of palliative care ([Box 1](#box1){ref-type="boxed-text"}) in the Caribbean and Africa should be viewed as an urgent public health problem. About 80% of cancer patients will have pain in the terminal phase of their disease \[[@pmed-0010005-b1]\], and we estimate that at least 25% of HIV/AIDS patients have substantial pain during their illness. ::: {.caption} ###### Box 1. The WHO Definition of Palliative Care ::: The WHO has defined palliative care as an approach that improves the quality of life for patients and their families facing the problems associated with life-threatening illness, through the prevention and relief of suffering. This is done through early identification, careful assessment, and treatment of pain and other problems---physical, psychological, and spiritual. Dying is regarded as a normal process, and death is neither hastened nor postponed \[[@pmed-0010005-b2]\]. The philosophy of hospice and palliative care acknowledges death, dying, and bereavement as a reality of life. Effective and relatively cheap methods exist for controlling pain and other symptoms. For example, the World Health Organization (WHO) has outlined a relatively cheap way of relieving cancer pain in about 90% of patients, which could be extended to patients with HIV/AIDS \[[@pmed-0010005-b2]\]. Sadly, most people in Africa and the Caribbean who need pain relief aren\'t receiving it \[[@pmed-0010005-b10]\]. Assessing Patients\' Needs {#s4} ========================== Several studies in East Africa have looked at the experience of dying, the quality of care at the end of life, and patients\' unmet needs \[[@pmed-0010005-b2],[@pmed-0010005-b11],[@pmed-0010005-b12]\]. Recurring themes are (1) unmet physical needs, including the need for relief of pain and other symptoms, (2) the need for food, (3) the high cost or unavailability of appropriate analgesic drugs, (4) the severe financial constraints on the family and caregivers, (5) the need for training of family caregivers, (6) lack of psychosocial support, and (7) social isolation due to the stigma attached to a diagnosis of HIV/AIDS. In the Caribbean, patients\' needs at the end of life appear to be similar to those of patients in many East African countries. A qualitative study in Grenada, in the Eastern Caribbean, showed that people preferred to die at home rather than in hospital and---in the absence of pain relief and much-needed counseling, information, and financial support---they took solace in spiritual comfort \[[@pmed-0010005-b13]\]. In Jamaica ([Box 2](#box2){ref-type="boxed-text"}), although data are scarce, it seems that patients\' needs are very similar to those in Grenada. Christianity is the principal religion of Jamaica, and faith in God and family support are critical factors in patient care at the end of life. Outside of the hospital setting, appropriate analgesics are difficult to access and are often unaffordable. Patients and caregivers are not provided with enough information to help them understand disease processes, and what to expect as the ill person nears death. There is little or no palliative care provision for patients with HIV/AIDS. ::: {.caption} ###### Box 2. Dying in Jamaica---A Typical Case Scenario ::: This fictional case scenario gives an impression of the sorts of problems that patients face at the Hope Institute, Kingston---Jamaica\'s first public hospice. A 50-year-old woman is diagnosed with inoperable lung cancer. Because of brachial plexus involvement, she experiences severe pain and weakness of her arm. She is treated at Kingston Public Hospital with palliative radiotherapy, which helps the pain for a few months. But then the pain returns, and she requires a high dose of slow-release morphine for pain control. She lives in the mountains, and her house is a two-and-a-half-hour bus ride from Kingston, the capital city. Unfortunately, the public pharmacy in Kingston is unwilling to dispense more than a week\'s supply of morphine at any one time, because they have limited supplies (there is a shortage of the drug in Jamaica) and because they think the patient\'s dose is unacceptably high. So she has to make the exhausting five-hour round trip every week. Her husband\'s health has also recently declined, and the woman\'s sister now has to care for the patient and her husband. The family now has the financial means to afford only one small meal a day, and they rely on donations from their church community in order to survive. Because the family\'s savings dwindle, and the public pharmacy faces further shortages of morphine, the woman with cancer requires multiple admissions to the hospice in Kingston over the last six months of her life in order to get suitable analgesia. Uganda\'s Public Health Approach {#s5} ================================ Uganda has made palliative care for patients with AIDS and cancer a priority in its National Health Plan \[[@pmed-0010005-b10]\]. In 1993, after conducting a feasibility study, Hospice Uganda was established in Kampala, making palliative care available to a population of about 2 million people (Uganda\'s population is 22 million people). There are now two other hospices, one in Mbarara serving 1 million people, and one in Hoima serving 350,000 \[[@pmed-0010005-b8]\]. The hospice care provided by these units is all home-based care. This type of care provision is designed to meet the cultural and practical needs of the people in Uganda, where most people prefer to die in their own homes, and where people are often buried in their household gardens. Hospice Uganda provides community-based care principally to patients suffering from HIV/AIDS and cancer. Almost all patients coming to the hospice have pain, and a great deal of attention is focused on good pain management. Uganda is only the third African country to have made morphine available and affordable to its patient population. Because of the dearth of legal prescribers (doctors, dentists, and vets only), in May 2004, Uganda changed the statute. This allowed midwives to prescribe pethidine, and allowed clinical palliative care nurses and clinical officers who are specially trained and registered to prescribe morphine. How was Uganda---an African country with a relatively under-funded health service---able to provide a palliative care service? A national program using a public health approach to reach those in need was established following principles outlined in the WHO\'s National Cancer Control Guidelines \[[@pmed-0010005-b4]\]. These guidelines outline the importance of assessing the magnitude of the problem, setting measurable objectives, evaluating possible strategies, and choosing priorities for initial activities. A series of workshops were held in Uganda between 1998 and 2000, where the WHO\'s "little cost, big effect" measures began to be addressed. The three key measures involve education, increased drug availability, and changes in government policy ([Figure 1](#pmed-0010005-g002){ref-type="fig"}). ::: {#pmed-0010005-g002 .fig} Figure 1 ::: {.caption} ###### The WHO\'s Triangle of Foundation Measures (Adapted with permission from "A Clinical Guide on Supportive and Palliative Care for People with HIV/AIDS" \[<http://hab.hrsa.gov/tools/palliative/>\]) ::: ![](pmed.0010005.g002) ::: Other African Initiatives {#s6} ========================= Four other African countries---Botswana, Ethiopia, Tanzania, and Zimbabwe---have made the development of home-based care a priority in dealing with the HIV/AIDS epidemic \[[@pmed-0010005-b2]\]. Botswana has an operational home-based care program integrated into its national health system, while in the other three countries, care is largely provided through private organizations. But few of the home-based care services in these countries include the capacity for providing effective pain relief \[[@pmed-0010005-b2]\]. The Next Steps {#s7} ============== By using strategies such as providing access to an essential short list of relatively cheap generic medications, and other methods recommended by WHO, it has now been proven that palliative care in the African context is affordable and achievable \[[@pmed-0010005-b2],[@pmed-0010005-b7],[@pmed-0010005-b14]\]. We believe that, following the Ugandan and Botswanan models, palliative care should be integrated into national government strategies. In order to begin to show governments the importance and economic justification for developing a palliative care health policy, it is clear that needs assessments are an essential first step. It is likely to be much less expensive to provide community-based care with family and community support at the end of life than to burden already overcrowded hospital wards with patients suffering end-stage disease. There is a long tradition, both in Africa and in the Caribbean, of caring for the disabled, the mentally ill, and the young and elderly sick at home. Both start-up and sustainable funding are enormous issues that will need to be addressed by local governments, international funding agencies, and charitable bodies. Advocating palliative care to decision makers, providing training programs for health professionals, and making medications available and affordable are important challenges. Research in individual countries is needed to assess whether the above recommendations are suitable locally. Hospice Africa Uganda is advocating to other African governments and assessing other African countries where local laws and customs may dictate the most suitable way to provide palliative care together with government support. Partnerships and a public health approach to palliative care must be the way forward. ::: {.caption} ###### Palliative Care Resources for the Developing World ::: African Palliative Care Association Representing Kenya, South Africa, Tanzania, Uganda and Zimbabwe E-mail: <apca@hospiceafrica.or.ug> Hospice Africa (Uganda) Resource and Training Centre PO Box 7757, Kampala, Uganda Tel: +256 41 266 867 / 510089; Fax: +256 41 510087---residence E-mail: <info@hospiceafrica.or.ug>; E-mail: <anne@hospiceafrica.or.ug> Centre for Palliative Learning Hospice Association of the Witwatersrand PO Box 87600, Houghton, Johannesburg 2041, South Africa Hospice Information At <http://www.hospiceinformation.info>. Click on "Training" to search for courses and conferences in palliative care and bereavement. Requires member\'s password to access this part of the website but membership is free to people in developing countries---contact hospice information at + 44 (0)870 903 3 903 (telephone), + 44 (0)20 8776 9345 (fax), or info@hospiceinformation. Information is also circulated quarterly by E-mail to members under the title of e-Choices. Palliative Care in Resource-Poor Settings A freely available overview of HIV/AIDS palliative care, written by Kathleen Foley, Felicity Aulino, and Jan Stjernswärd. At <http://hab.hrsa.gov/tools/palliative/chap19.html>. Living Well with HIV/AIDS A freely available manual on nutritional care and support for people with HIV/AIDS, by the Food and Agriculture Organization of the United Nations. At <http://www.fao.org/DOCREP/005/Y4168E/Y4168E00.HTM>. Cancer Pain Relief: A Guide to Opioid Availability A section of this guide, by the WHO, is freely available at <http://www.medsch.wisc.edu/painpolicy/publicat/cprguid.htm>. Citation: Spence D, Merriman A, Binagwaho A (2004) Palliative care in Africa and the Caribbean. PLoS Med 1(1): e5. WHO : World Health Organization [^1]: Dingle Spence is an associate lecturer at the University Hospital of the West Indies and the director of the Hope Institute, Kingston, Jamaica. Anne Merriman is the medical director of Hospice Africa Uganda, Kampala, Uganda. Agnes Binagwaho is the executive secretary of the National AIDS Control Commission, Kigali, Rwanda. [^2]: Competing Interests: DS declares that she has no competing interests. AM and AB are on the editorial board of PLoS Medicine.	PubMed Central	2024-06-05T03:55:48.578095	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523828/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e5", "authors": [ { "first": "Dingle", "last": "Spence" }, { "first": "Anne", "last": "Merriman" }, { "first": "Agnes", "last": "Binagwaho" } ] }
PMC523829	In wealthy countries, state-funded research has yielded breakthroughs in molecular biology, chemistry, and engineering. These advances have been taken up by the pharmaceutical industry and applied to drug development for a growing range of illnesses and conditions. As a result, patients have access to new drugs that are better tolerated, more specific, and more effective than old ones. In poor countries, however, millions of people have yet to experience the benefits wrought by science. The deadly infectious diseases that plague them, such as sleeping sickness, Chagas disease, and visceral leishmaniasis, fail to arouse the interest of drug developers. The Drugs for Neglected Diseases Initiative (DNDi) is a new, not-for-profit organisation set up to correct this fatal imbalance by developing new drugs for these forgotten patients. Dropped off the Radar Screen {#s2} ============================ Most of the drugs still used to treat 'neglected diseases' were developed in colonial times. These are often expensive, difficult to administer, and hard to tolerate; several of them are also becoming ineffective because of increasing parasite resistance. Very few new alternatives have been developed in the past decades: between 1975 and 1999, 1,393 new drugs were made available to the public, but only 16 of these were meant for neglected diseases \[[@pmed-0010006-b1]\]. What makes the lack of drugs more difficult to accept is that scientists know an enormous amount about kinetoplastids, the organisms responsible for sleeping sickness, Chagas disease, and leishmaniasis \[[@pmed-0010006-b2]\]. The wealth of knowledge generated in this field could easily be used for drug development if the treatment of neglected diseases were perceived as financially attractive. But populations affected by neglected diseases have no purchasing power, so there is no financial incentive for drug companies to develop the drugs. The basic mechanics of the market-driven system are failing to help these populations. So most scientific research stops at the publication stage or falls through the gaps at different stages of the drug development pipeline ([Figure 1](#pmed-0010006-g001){ref-type="fig"}) \[[@pmed-0010006-b3]\]. ::: {#pmed-0010006-g001 .fig} Figure 1 ::: {.caption} ###### The Drug Development Pipeline Because of the gaps in the development pipeline, potential new drugs for neglected diseases often stay stuck at an early stage of development. (Photos: World Health Organization/P.Virot and World Health Organization/Eric Miller) ::: ![](pmed.0010006.g001) ::: Whose Role Is It, Anyway? {#s3} ========================= It is dangerous to oversimplify the causes of this situation. What share of responsibility for the world\'s health is borne by the pharmaceutical industry, which has the know-how and the resources for innovation? Aren\'t international organisations also partly responsible? After all, they are the ones who allocate major funding for health programmes and encourage research programmes. And what about public research institutions in rich countries that generate the knowledge used by industry? Governments have the power to influence their research priorities and drug development decisions, either through funding or direct involvement. Unfortunately drugs for neglected diseases are low priority for governments \[[@pmed-0010006-b4]\]. They tend to prioritise research with potential commercial applications instead. Responses to the Crisis {#s4} ======================= All is not doom and gloom. In the past few years, there has been some movement on research and development (R&D) for neglected diseases. Despite its broad mandate and limited resources, the Special Programme for Research and Training in Tropical Diseases (TDR)---established and funded by the World Health Organization, the World Bank, and the United Nations Development Programme---can be credited with several important successes in the fight against malaria and leishmaniasis. The Medicines for Malaria Venture and the Global Alliance for TB Drug Development were set up as public--private partnerships to tackle malaria and tuberculosis. These partnerships were made possible by the fact that malaria and tuberculosis are global diseases, affecting patients in the North and South, so there was enough of a market to persuade industry to develop new drugs for these diseases. A different solution, however, was needed for diseases that are limited to tropical countries, are of no military or strategic interest to wealthy countries, and are not supported by markets or patients\' organisations capable of attracting the attention of politicians. This is the kind of solution put forward by the DNDi. A Collaborative Not-for-Profit {#s5} ============================== DNDi is a not-for-profit organisation designed to mobilise resources for R&D of new drugs for neglected diseases. Many people and organisations around the world share an ambition to redress the lack of new treatments for neglected diseases, and bring the benefits of science to forgotten patients. Several of them came together to create DNDi: one humanitarian organisation---Médecins Sans Frontières; five research institutions---the Oswaldo Cruz Foundation from Brazil, the Indian Council for Medical Research, the Kenya Medical Research Institute, the Ministry of Health Malaysia, and the Pasteur Institute from France; and the TDR ([Box 1](#box1){ref-type="boxed-text"}). ::: {.caption} ###### Box 1. From Pipeline to Patients---Some Key Organizations ::: DNDi: <http://www.dndi.org> TDR: <http://www.who.int/tdr> Medicines for Malaria Venture: <http://www.mmv.org> Global Alliance for TB Drug Development: <http://www.tballiance.org> Oswaldo Cruz Foundation: <http://www.fiocruz.br> Indian Council for Medical Research: <http://icmr.nic.in/home.htm> Kenya Medical Research Institute: <http://www.kemri.org> Ministry of Health Malaysia: <http://dph.gov.my/> Pasteur Institute: <http://www.pasteur.fr/externe> Médecins Sans Frontières: <http://www.msf.org> The initiative is a virtual organisation with a growing network of academic and R&D expertise at its disposal. The different players involved in DNDi are bringing their knowhow in parasitology and clinical trials, their experience treating neglected patients, and their drug manufacturing capacity. They are pooling these resources to move drugs stuck in the pipeline all the way to the patients themselves. Pharmaceutical companies have a particularly important role to play: they possess vast repositories of molecules, the means to move from development to industrial production, and highly specialised teams of researchers. Their contribution will be crucial to the success of DNDi. Matching Needs and Opportunities {#s6} ================================ DNDi is a needs-driven initiative---in other words, the needs of patients suffering from neglected diseases are paramount in its search for new drugs to treat them. The organisations that make up DNDi have firsthand knowledge of these needs because they work with patients in disease-endemic countries throughout the developing world. The initiative is taking this knowledge of patient needs, matching it with opportunities in R&D, and pushing the most relevant projects through the pipeline. Ultimately, neglected patients will have access to drugs targeting their specific diseases, drugs that were designed with them specifically in mind---such as short-course, low-toxicity treatments that don\'t require hospitalisation, or tablets to swallow rather than injections. To identify opportunities in R&D that are both relevant to patient needs and that meet required criteria of scientific merit, DNDi is sending out calls for letters of interest to the scientific community via advertisements in journals and posted on the DNDi website (<http://www.dndi.org>). These have already pinpointed several promising projects. DNDi is also proactively contacting scientists working on infectious diseases, and surveying published literature for research of interest. In the Pipeline {#s7} =============== DNDi\'s project portfolio currently holds nine projects at different stages of development to address identified needs for the treatment of visceral leishmaniasis, sleeping sickness ([Box 2](#box2){ref-type="boxed-text"}), Chagas disease, and malaria ([Figure 2](#pmed-0010006-g002){ref-type="fig"}). At discovery stage, DNDi is working on validating the kinetoplastid enzyme dihydrofolate reductase as a potential target for leishmaniasis, trypanosomiasis, and Chagas disease, and on identifying inhibitors of the kinetoplastid enzymes trypanothione reductase and protein farnesyltransferase. It is also conducting high throughput screening on whole cell trypanosomes to discover novel lead compounds. ::: {.caption} ###### Box 2. New Drugs for Sleeping Sickness ::: Only a few drugs exist to treat sleeping sickness, and they are toxic or difficult to administer. Melarsoprol kills one in 20 patients. Eflornithine requires four daily infusions over 14 days. Given these limited options, DNDi is focusing on identifying new compounds that can cross the blood--brain barrier to treat second stage sleeping sickness. DNDi is using high throughput screening on whole cell trypanosomes to discover novel lead compounds, and is working to identify and optimise inhibitors of the enzyme protein farnesyltransferase. The initiative is working on validating the kinetoplastid enzyme dihydrofolate reductase as a drug target. Identifying trypanothione inhibitors is also relevant to other trypanosome parasites. These are long term projects. Nifurtimox, a drug for Chagas disease, has been used to treat sleeping sickness since the 1970s in some isolated places. It has never been extended to more people because no one has studied its safety or effectiveness. DNDi will assess its short-term usefulness by conducting clinical trials on a treatment combination of eflornithine and nifurtimox. DNDi will continue to explore other short- and medium-term projects. ::: {#pmed-0010006-g002 .fig} Figure 2 ::: {.caption} ###### DNDi Projects DNDi\'s project portfolio contains nine projects spread out across the drug development pipeline for the treatment of leishmaniasis, sleeping sickness, Chagas disease, and malaria. HAT, human African trypanosomiasis (sleeping sickness); VL, visceral leishmaniasis. ::: ![](pmed.0010006.g002) ::: The R&D of new drugs is time-consuming and expensive if the process starts at the early discovery stage, because of the associated risk of project attrition along the way. DNDi is therefore investing resources in several pre-development and development projects as well. These include developing fixed dose combinations of artesunate/amodiaquine and artesunate/mefloquine for use against chloroquine-resistant malaria in Africa and Asia, respectively; pushing for registration of paromomycin for use against visceral leishmaniasis in Africa; assessing combinations of existing drugs for visceral leishmaniasis; and evaluating the usefulness of nifurtimox in combination with eflornithine in the treatment of sleeping sickness. Advocacy for Change {#s8} =================== Governments can---some might say should---influence drug development choices. DNDi strongly believes that governments in both developed and developing countries should take an active interest in the R&D of new drugs for neglected diseases. In parallel to its own drug development activities, DNDi is working to raise awareness of the neglected disease crisis among key policy- and decision-makers, for instance the European Commission and the National Institutes for Health in the United States. Conclusion {#s9} ========== In the poorer countries in the world, over 350 million people are at risk from neglected diseases. Currently available treatments are inadequate or nonexistent, and new solutions are urgently needed. DNDi is working to ensure that the advances of science that have brought health and comfort to wealthy nations also benefit these neglected populations. Citation: Pécoul B (2004) New drugs for neglected diseases---From pipeline to patients. PLoS Med 1(1): e6. DNDi : Drugs for Neglected Diseases Initiative R&D : research and development TDR : UNICEF/UNDP/World Bank/World Health Organization Special Programme for Research and Training in Tropical Diseases WHO : World Health Organization [^1]: The author declares that he has no competing interests. [^2]: Bernard Pécoul is the Executive Director of the Drugs for Neglected Diseases Initiative (DNDi), Geneva, Switzerland. E-mail: <dndi@dndi.org>	PubMed Central	2024-06-05T03:55:48.579600	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523829/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e6", "authors": [ { "first": "Bernard", "last": "Pécoul" } ] }
PMC523831	About a decade ago, I went wandering around Cairo\'s City of the Dead. This might sound like a grim bit of tourism, but my connection to that vast necropolis runs deep---quite literally, as my family is buried there. After visiting their grave, I rambled through the city\'s dusty alleyways, past hundreds of years of history. Yet what I remember most about that day was not one of the many magnificent tombs, but a simple brick building with a sign, of all things, for a family planning clinic. I was certainly surprised by my discovery; in retrospect, I should not have been. That part of Cairo is home to hundreds of thousands of people for whom looking after the dead is a way of life. Their fertility invigorates the funereal air: the caretaker of my family\'s tomb, for example, had a blooming family of his own living near the grave. Where better to offer family planning than in a place so poor that reproduction seemed more a matter of fate than choice? The Cairo Conference {#s2} ==================== That visit is a fitting metaphor for the field of reproductive health as a whole. Ten years ago, officials, experts, and activists from 179 countries also came to Cairo for the International Conference on Population and Development (ICPD). The conference produced a 20-year plan of action that focused on universal access to reproductive health services, including family planning and sexual health; reducing infant, child, and maternal mortality; better education, especially for girls; equality between men and women; and sustainable development. The ICPD\'s key achievement was to reorient thinking on reproduction away from narrowly defined, government-dictated population control to a broader appreciation of reproductive and sexual well-being within health care systems, a view driven by individual choice and rights, not official priorities. "The Cairo Conference was a peak moment," says Sally Ethelston, vice president for communications at Population Action International, one member of a consortium of non-governmental organisations launching a report card to mark the anniversary of the Cairo Conference in early September. "There were times when people were excited that they had accomplished something, and you could see it on their faces." Today, however, the mood is very different. While progress has been made on some of the plan\'s targets, effort has faltered on others. And the conference "camaraderie" that Ethelston describes has given way to conflict between faith and science, over abortion and condoms. Like signs of life in the City of the Dead, the Cairo Conference gave birth to great expectations, some of which have already expired. Baby Steps Towards Cairo\'s Goals {#s3} ================================= So, how far has the developing world come towards meeting the ICPD goals? There has certainly been progress on institutional reform in some countries, according to a recent survey of national policies by the United Nations Population Fund (UNFPA) \[[@pmed-0010009-b1]\]. For example, more than a third of the 151 countries questioned have introduced legislation on reproductive rights, and almost half have expanded their primary health care services to include family planning.[](#pmed-0010009-g001){ref-type="fig"} ::: {#pmed-0010009-g001 .fig} ::: {.caption} ###### The birth of a baby on August 15, 2000, brought India\'s population to one billion (Photo: Raghu Rai, on behalf of the David and Lucile Packard Foundation) ::: ![](pmed.0010009.g001) ::: But translating policy into action has been difficult. Overall, the picture is one of patchy success, according to Susheela Singh, director of research at The Alan Guttmacher Institute, a nongovernmental research organisation. Official statistics, as limited as they are for many aspects of reproductive and sexual health, show mixed results. On a positive note, global population growth has slowed to roughly 77 million people a year \[[@pmed-0010009-b2]\]. But while fertility rates have fallen in some developing countries, such as Mexico, they remain stubbornly high in others, such as Ethiopia \[[@pmed-0010009-b3]\]. Over the past decade, contraceptive use has grown, but so has demand, and there are now an estimated 201 million women in developing countries whose need for modern birth control goes unmet, resulting in 60 million unintended pregnancies a year \[[@pmed-0010009-b4]\]. Progress on legalising abortion has been slow, and an estimated 19 million abortions a year still occur under unsafe conditions \[[@pmed-0010009-b5]\]. Despite growing awareness of sexually transmitted disease, the annual number of sexually transmitted infections remains worryingly high at 340 million worldwide \[[@pmed-0010009-b6]\]. While infant mortality rates have improved somewhat, maternal mortality figures have barely budged. An estimated 529,000 women still die every year from complications of pregnancy and childbirth. The highest rates are in sub-Saharan Africa, where, on average, 920 women die for every 100,000 live births, compared with 24 deaths per 100,000 live births in Europe \[[@pmed-0010009-b7]\]. This is all the more distressing, says Vivien Tsu, senior programme officer at the Program for Appropriate Technology in Health, because these women\'s lives could be saved through straightforward measures and basic technologies, such as access to skilled midwives, simple drugs like magnesium sulphate for eclampsia and oxytocin for post-partum bleeding, cellular phones to call for help, and transportation to emergency obstetric centres. Obstacles to Reproductive Health {#s4} ================================ So why hasn\'t more been achieved? One problem is certainly money. The 1994 Cairo Conference estimated the cost of implementing programmes for family planning, maternal health, and prevention of sexually transmitted diseases, as well as data collection and analysis in developing countries, at \$18.5 billion by 2005---or \$24.3 billion in today\'s dollars. The goal was to mobilise one-third of the money from donor nations, and the rest from developing countries themselves \[[@pmed-0010009-b8]\]. Last year, global spending on reproductive health and services reached \$14.7 billion, according to estimates from UNFPA, the Joint United Nations Programme on HIV/AIDS, and the Netherlands Interdisciplinary Demographic Institute \[[@pmed-0010009-b8]\]. Encouragingly, investment has increased since 2001, when the momentum of ICPD seemed to falter and international spending fell to \$9 billion. But this is still wide of the mark. While developing countries have failed to meet their conference commitments, it is donor countries that are most remiss: rich country contributions reached an estimated \$2.3 billion in 2003 \[[@pmed-0010009-b8]\], a far cry from the conference target of \$6.1 billion (or \$8.1 billion in today\'s dollars) by 2005. Reproductive health is not alone in waiting for donors to give generously. For all the rhetoric at international summits, few rich countries have lived up to their lofty pledges of debt relief and of dedicating 0.7% of their gross domestic product to overseas development assistance. But as Steve Sinding, head of the International Planned Parenthood Federation (IPPF), points out, there are other reasons too for the shortfall. In the past donor interest was largely stimulated by fears of a population crisis. When the Cairo Conference reframed issues in terms of women\'s health and reproductive rights, rather than an impending population explosion, Sinding argues, the "demographic rationale" was lost, taking funding with it. Moreover, there are other issues competing for international funding, most notably AIDS. At the time of the Cairo Conference, 20 million people were infected with HIV; today the number has grown to an estimated 38 million \[[@pmed-0010009-b9]\]. AIDS threatens to derail the Cairo Conference plan of action. Through maternal-to-child transmission, and wide-scale orphaning, HIV threatens to reverse small successes at reducing infant and child mortality. By killing off teachers and sapping household incomes, AIDS is sabotaging education. By killing off scarce medical workers and overwhelming fragile health care systems, the disease is compromising reproductive health services. Gender equity is undermined, as women and girls bear the brunt of the epidemic, as caregivers, breadwinners, or patients themselves. Roughly half of the money spent on reproductive health last year went towards HIV/AIDS. And billions more is on the way, from the likes of the Global Fund to Fight AIDS, Tuberculosis, and Malaria and the United States President\'s Emergency Plan for AIDS Relief, which promises \$15 billion over five years to HIV/AIDS programmes \[[@pmed-0010009-b10]\]. But much of this money is going into AIDS-specific programmes that do not address reproductive health more broadly. Even as the world is gearing up to scale up AIDS prevention and treatment to millions worldwide, few of the agencies involved come from the world of reproductive and sexual health. This is a pity because it means that HIV/AIDS programmes are not making use of valuable infrastructure and expertise already on the ground in places where AIDS hits hardest. Given that 57% of HIV infections in sub-Saharan Africa are among women \[[@pmed-0010009-b9]\], and that, for many of them, family planning clinics are their sole contact with the formal health care system, it seems odd not to integrate such services into the wider battle against HIV. Such centres can offer not only HIV testing and counselling, as well as condoms (against the double whammy of unwanted pregnancy and HIV infection), but also a broad-based message of good sexual health that can help protect against HIV and other sexually transmitted diseases. Moreover, pre- and ante-natal care provide an opportunity to stop mother-to-child transmission of HIV in its tracks.[](#pmed-0010009-g002){ref-type="fig"} ::: {#pmed-0010009-g002 .fig} ::: {.caption} ###### Condom distribution in Soweto, South Africa (Photo: Arjen van de Merwe, Population Concern) ::: ![](pmed.0010009.g002) ::: Where once family planning was the darling of international donors, HIV is now the cause célèbre. "There\'s a lot of resentment about the spotlight moving on," says Kevin O\'Reilly, a former reproductive health specialist now at the department of HIV/AIDS at the World Health Organization. However, there are now attempts to bring the two together. Meetings earlier this year in Switzerland, New York, and Bangkok have led to calls to action to strengthen links between programmes addressing HIV/AIDS and sexual and reproductive health. While this should help in the battle against AIDS, the money which flows to AIDS should also benefit reproductive health. Ideological Battles {#s5} =================== Arguably the most formidable obstacle to that union, and indeed further progress in improving reproductive health, is ideology. Since the Cairo Conference, a fierce battle has emerged between religious conservatives who eschew abortion and condoms in favour of abstinence and fidelity, and more liberal voices who argue for a full armamentarium to tackle these problems. The clash is loudest in the field of HIV/AIDS, where the President\'s Emergency Plan for AIDS Relief allocates a third of its funding for disease prevention to programmes focusing on abstinence and fidelity; public health experts argue that such an approach is ineffective at best, and dangerous at worst, without an equal emphasis on the availability of condoms for all. But the clash resounds in the wider arena of reproductive health as well. Four years ago, the ICPD\'s central target---access to reproductive services for all by 2015---failed to make it into the Millennium Development Goals, largely because of political nervousness. But as Kofi Annan, United Nations secretary-general, has pointed out, progress on the other key targets, such as eradication of poverty and hunger, will not be achieved without a focus on women\'s rights, education, reproductive health, and family planning. The fight between conservatives and liberals is clearest in the case of the US, which is the world\'s leading bilateral donor on reproductive health, spending \$429 million this year \[[@pmed-0010009-b11]\]. However, this money comes with strings attached, says Françoise Girard, a reproductive rights lawyer in New York. Some of these are subtle. For example, Girard points to American pressure on several Asian and Latin American governments---during recent regional meetings to mark the anniversary of the Cairo Conference---not to re-affirm their commitment to the ICPD plan of action, with its emphasis on a full suite of reproductive rights and services. Other strings are more obvious. In 2001, George W. Bush reinstated the Mexico City Policy, otherwise known as the "Global Gag Rule", which denies US family planning assistance---including money and contraceptive supplies---to any non-American group unless it certifies that it neither performs nor endorses abortion. IPPF, Marie Stopes International, and their local affiliates have been hard hit by the Rule, scaling back services in Kenya, Ghana, and elsewhere that offered essential health care to thousands of women and children. Then there is the Kemp-Kasten Amendment, a piece of US legislation which prohibits US assistance to any organisation as deemed by the President that "supports or participates in the management of a program of coercive abortion or involuntary sterilization." At the behest of conservative supporters, President Bush has used the amendment to withhold \$34 million in annual congressional appropriations to the UNFPA for the past three years. The UNFPA says that the \$34 million could have been used to prevent 2 million unintended pregnancies, 800,000 induced abortions, 4,700 maternal deaths, and 77,000 infant and child deaths. The White House accuses UNFPA of abetting coercive reproductive practices in China---a claim which the UNFPA strenuously denies. Moreover, a number of international delegations, including one from the US State Department in 2002, have investigated the UNFPA\'s activities in China and failed to find evidence to support such allegations. Fortunately, other donors are stepping in to fill the breach: earlier this year, for example, the United Kingdom announced it would raise its contribution to the UNFPA to £80 million over the next four years, as well as increase its support to IPPF by a third. But even if the shortfall is made up, the ill will such clashes have engendered cannot be so easily salved. A Call for Strong Leadership {#s6} ============================ Getting it right on reproductive health cannot wait another decade. The largest generation of young people in history---a whopping 1.2 billion aged 10--19 years---is entering adulthood \[[@pmed-0010009-b1]\]. They are making their sexual debut at ever earlier ages, against a backdrop of rising sexually transmitted diseases and growing social conservatism, which makes clear information, frank discussion, and free choice on abortion, contraception, and sexual health extremely difficult. More than ever, reproductive health needs strong leaders in rich and poor countries alike to mobilise both money and political commitment. Reproduction is a sexy subject; it is time the world again paid it the attention it deserves. ::: {.caption} ###### Useful Links ::: The Cairo Conference: <http://www.iisd.ca/cairo.html> Population Action International: [www.popact.org](www.popact.org) UNFPA: [www.unfpa.org](www.unfpa.org) Program for Appropriate Technology in Health: [www.path.org](www.path.org) The Alan Guttmacher Institute: [www.guttmacher.org](www.guttmacher.org) The Joint United Nations Programme on AIDS: [www.unaids.org](www.unaids.org) Netherlands Interdisciplinary Demographic Institute: [www.nidi.nl](www.nidi.nl) IPPF: [www.ippf.org](www.ippf.org) Global Fund to Fight AIDS, Tuberculosis, and Malaria: [www.theglobalfund.org](www.theglobalfund.org) The World Health Organization HIV/AIDS Programme: [www.who.int/hiv/en](www.who.int/hiv/en) Citation: El Feki (2004) The birth of reproductive health: A difficult delivery. PLoS Med 1(1): e9. ICPD : International Conference on Population and Development IPPF : International Planned Parenthood Federation UNFPA : United Nations Population Fund [^1]: Shereen El Feki is the healthcare correspondent at The Economist magazine, London, United Kingdom. E-mail: <Shereenelfeki@economist.com> [^2]: Competing Interests: The author declares that she has no competing interests.	PubMed Central	2024-06-05T03:55:48.580764	2004-9-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523831/", "journal": "PLoS Med. 2004 Oct 7; 1(1):e9", "authors": [ { "first": "Shereen", "last": "El Feki" } ] }
PMC523832	As an intern, I took care of the first patients with HIV/AIDS at San Francisco General Hospital, and so I grew up with AIDS in the early days of my medical career. We struggled through the confusion about what was making people so sick, and each new day brought a new discovery about the disease and its consequences. I went through that evolutionary process along with everybody else, and it shaped me in many profound ways. Before long, I recognized that this wasn\'t a disease of "those people over there." This was a disease that could strike anyone, anytime. And as physicians, we had to adjust our thinking about our own vulnerability to occupational risk, and to emphasize prevention, because there wasn\'t going to be a cure for a long, long while. And not only physicians had to rethink things---AIDS has reshaped society\'s very notions of the most basic human behaviors. I was in Uganda in 1985, early in the AIDS epidemic there. We knew then where that epidemic was going to go, absent an effective vaccine or cure, but few of us could have imagined that it would evolve so quickly without an end in sight. While the people of Africa have achieved a huge amount in tackling HIV/AIDS, particularly in Uganda, the epidemic is far from being under control on that continent and is spreading through other parts of the world with alarming speed. The Crisis of Human Resources {#s2} ============================= The theme of this year\'s International AIDS Conference in Bangkok was "Access for All." Over the past few years, it has become increasingly apparent that a critical component of assuring access to care and treatment is human capital. Like fiscal capital, human resources are essential to ending the AIDS pandemic. I visited Africa with US Health and Human Services Secretary Tommy Thompson and many AIDS experts last December, and we saw evidence of this critical need in every country we visited. The miracles of modern science are meaningless without systems and people to deliver them to those in need.[](#pmed-0010010-g001){ref-type="fig"} ::: {#pmed-0010010-g001 .fig} Figure 1 ::: {.caption} ###### Estimated Percentage of Adults in Need of Antiretroviral Treatment Who Are Receiving It, as of March of 2004 (This graphic is based on an image by the World Health Organization, available at <http://www.who.int/3by5/en/coverage_march2004.jpg>) ::: ![](pmed.0010010.g001) ::: The World Health Organization estimates that of the 40 million people worldwide infected with HIV, 6 million need immediate, life-sustaining antiretroviral therapy. Fewer than 400,000 people in developing countries have access to such treatment ([Figure 1](#pmed-0010010-g001){ref-type="fig"}) \[[@pmed-0010010-b1],[@pmed-0010010-b2]\]. There are too few skilled health care workers to provide reliable delivery and administration of these life-saving therapies. According to a recent Institute of Medicine report, and a study sponsored by the US Agency for International Development, the number of health care workers in many African countries is actually shrinking as they are lured to developed countries by better pay and professional opportunities ([Box 1](#box1){ref-type="boxed-text"}) \[[@pmed-0010010-b2],[@pmed-0010010-b3]\]. Reversing this brain drain is essential over the long-term, as HIV treatment and care will be required for decades. In the short-term, the Institute of Medicine called for expanded efforts "to bring qualified volunteer initiative medical professionals into both urban and rural areas to support prevention, care, and training programs" \[[@pmed-0010010-b2]\]. I could not agree more that addressing the human resource needs will be essential as we move forward---and not just for HIV/AIDS programs, but for all aspects of public health and health care.[](#box1){ref-type="boxed-text"} ::: {.caption} ###### Box 1. The Brain Drain: Facts and Figures \[[@pmed-0010010-b2],[@pmed-0010010-b3]\] ::: Only 360 of the 1,200 doctors trained in Zimbabwe during the 1990s continue to practice within the country.Two-thirds of University of Zimbabwe medical students intend to leave the country after graduating, and one of the country\'s major 1,000-bed teaching hospitals lacks a single qualified pharmacist.In Zambia, only 50 of the 600 doctors trained locally since independence have remained in the country.In Ghana, 320 nurses are recorded to have been lost in 1999, roughly equivalent to that country\'s annual output of nurses; losses for the year 2000 totaled 600. It has now been shown, beyond any doubt, that even in resource-poor countries with the most basic health infrastructure, people get the same benefit from treatment and prevention interventions as those in the rich world \[[@pmed-0010010-b4]\]. In fact, surveys in Cape Town, Kampala, Khayelitsha, and Senegal found rates of adherence to antiretroviral therapy of 90%--94%, compared with estimates of 70% in developed countries \[[@pmed-0010010-b5],[@pmed-0010010-b6],[@pmed-0010010-b7]\]. When You Have Seen the Faces {#s3} ============================ We hear the numbers---the millions upon millions infected---and we grow numb. That is why we must go to the front lines---the households and communities---and start focusing on each individual living with HIV. I was at the dedication of an AIDS clinic in Kenya. It was raining, and we were waiting outside with our umbrellas. A 12-year-old girl in front of me turned around and leaned her head against my belly and said, "Could you take me to America? I need drugs." If you take that girl\'s face and multiply it a thousand times---that is the memory I bring home from Africa: the faces of the children and their asking, "Why are so many of our parents dying? Why are we dying?" We visited a US Centers for Disease Control and Prevention (CDC) program in the very remote areas of Uganda where there are no roads and it is impossible for people to come into population centers to receive HIV testing and other services. Young staff from the CDC are working with Ugandans and community organizations in that area to deliver antiretroviral therapy. Some may think that the difficulties of delivering antiretroviral therapy into such a remote area are overwhelming---and some may question whether this is a sustainable intervention. But once you see firsthand what miracles are possible, your world view changes almost overnight. What we saw was the success of a wonderful home-based treatment and care program for people who don\'t have access through other means. And when I say "home-based care," picture a hut without running water or electricity, where only motorcycles are available to deliver medications. The first step of the program is to provide clean water. Coliforms and other pathogens in the water supply for the household are removed through an inexpensive water vessel fitted with a filter and through a chlorination process. In addition, a cotrimoxazole tablet is given every day, which, in one patient\'s words, changed his life because he began to feel well almost immediately. Not only do the cotrimoxazole prophylaxis and the water treatment improve diarrheal illness, but malarial parasitemia also drops. So that is a very positive, unexpected consequence of just two very simple and inexpensive interventions. Many patients with HIV/AIDS in Africa also have tuberculosis and are put on tuberculosis therapy in addition to cotrimoxazole. As a result, they begin to feel better even before they begin antiretroviral therapy.[](#pmed-0010010-g002){ref-type="fig"} ::: {#pmed-0010010-g002 .fig} ::: {.caption} ###### Behavior change club at a technical school in Entebbe, Uganda (Photo: Arjen van de Merwe/World Population Foundation) ::: ![](pmed.0010010.g002) ::: We spent time with one of the patients in the home-based care program. As she began to participate in these programs, tests became available to measure her CD4 count. She explained to me what her CD4 count was, what it meant, and how it improved when she started the cotrimoxazole and tuberculosis therapy. She had begun taking three antiretroviral drugs and held up her pill box to explain her regimen in detail. Every week a Ugandan health aide delivered her supply of pills on a CDC motorcycle and monitored her adherence to the treatment. Not only was she extremely reliable in taking her medications, but she also knew more about them and their side effects than most of the patients I treated at San Francisco General. She was also an expert in HIV prevention. I asked her, "What do you do to protect your three young sons from this infection?" She replied, "Every day I take them by the hand, and I go out of the house and I say, 'Do you see that mound of dirt? That is your father's grave. Your father acquired this fatal infection through sex. Be careful.\'" And then she talks to them about the "ABCs" ("A" for abstinence, "B" for being faithful, "C" for condoms). So when you see a story like that unfold in the middle of Africa, it\'s impossible not to be hopeful. And yet, it\'s also very sobering because we are reminded of our responsibility. The question is not what the international health community is accomplishing in these countries now, but what we could accomplish if we joined together to really fight this war on AIDS. Such a story also inspires hope because you can see the multiplier effect that comes from taking on one problem and can see the way that effort can expand to encompass and address a much greater set of problems. Beyond ABCs---Diagnosis and Responsibility {#s4} ========================================== When we think about successful prevention models in Uganda, "ABC" certainly stands out \[[@pmed-0010010-b8]\]. However, at this point in the epidemic curve, other letters must also be considered. Most HIV transmission is accounted for by infected people having risky sex with uninfected people. Both in the US and in Africa, studies show that most infected people engaging in risky sex are unaware of their infection status, and that when their infection is diagnosed, they usually take steps to protect the others with whom they are having contact \[[@pmed-0010010-b9],[@pmed-0010010-b10],[@pmed-0010010-b11],[@pmed-0010010-b12]\]. So let\'s add the letter "D" for diagnosis. In fact, improving efforts to help people choose risk avoidance and to diagnose those who are already infected is the cornerstone of the CDC\'s new domestic HIV prevention strategy. Diagnosis is extremely important in many African communities, especially where the number of discordant couples---where one individual is infected and the other is not---is high. Sadly, many couples "being faithful" now do not realize that one partner is already infected and are not being reached with diagnostic testing programs. So "ABCD" is a concept that I would like to put out on the table as food for thought. Of course, there is another letter that we need to stress: the letter "R," for responsibility: personal sexual responsibility is a critical component of HIV prevention. Many women and girls become infected after being raped by men or because their social circumstances rob them of the power to refuse sex. Men must be held accountable for greater sexual responsibility and for ending sexual violence and degradation of women and girls. HIV prevention programs need to emphasize responsibility, but not lose sight of the fact that responsibility can be practiced only with personal autonomy, which many women and girls simply do not have. Expanding the Team to Meet the Needs {#s5} ==================================== The innovative programs and ideas emerging in Africa can change the picture of the AIDS epidemic. The purchase of antiretroviral drugs for Africans is not the big challenge. Access to drugs will improve in Africa. The real challenges are delivering drugs in a safe and effective way, monitoring therapy, and sustaining the pipeline of drugs so that ongoing treatment can be guaranteed. In the example of the home-based program in Uganda, we have seen that these challenges can be overcome. Expanding access to prevention, care, and treatment services isn\'t going to be easy, but it is certainly possible. It will take unprecedented commitment by people in the public sector, the private sector, faith communities, and community organizations, and perhaps most importantly, individual volunteers who make up their minds to contribute in any way they can. Last fall, the US Peace Corps announced that it was activating programs in some countries that allow volunteers to help communities fight the AIDS epidemic, but this is just one of many steps that are being taken. The US president\'s Emergency Plan for AIDS Relief will provide \$15 billion, including almost \$10 billion in new funds, over five years for international AIDS assistance \[[@pmed-0010010-b13]\], and I am part of the team that is charged with making this plan happen. I look forward to learning from others in the global health community how we can best expand our impact and collectively find a way to support the delivery of prevention messages and life-saving medications to everyone in Africa---and especially to that little girl at the Kenyan clinic who touched my heart. Citation: Gerberding J (2004) Steps on the critical path: Arresting HIV/AIDS in developing countries. PLoS Med 1(1): e10. CDC : Centers for Disease Control and Prevention [^1]: Julie Gerberding is the director of the Centers for Disease Control and Prevention, Atlanta, Georgia, United States. E-mail: <hhc7@cdc.gov> [^2]: Competing Interests: The author declares that she has no competing interests.	PubMed Central	2024-06-05T03:55:48.582585	2004-10-19	{ "license": "Creative Commons Zero - Public Domain - https://creativecommons.org/publicdomain/zero/1.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523832/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e10", "authors": [ { "first": "Julie", "last": "Gerberding" } ] }
PMC523834	In 1996, Richard Horton, editor of the Lancet, chastised much of current surgical research and, in particular, questioned the usefulness of the case series as a predominant form of communication among surgeons \[[@pmed-0010013-b1]\]. He asked a poignant question: "Does surgical research have a future?" Nearly a decade later, it is important for surgeons and non-surgeons alike to revisit Horton\'s challenge. Why Surgeons Favor Case Series {#s2} ============================== Randomized controlled trials (RCTs) have become the pillar of clinical research. Such trials attempt to obtain an unbiased randomization of patients with respect to known and unknown baseline conditions and to assess the effects of an intervention. However, only a minority of surgical studies involve a valid randomization scheme. The case series remains a favored method of clinical investigation in surgery. Case series are easy to perform, require less resources in terms of personnel and funds, can be performed at a single center, and, for many surgeons, represent a means to illustrate their surgical method and skills. In many instances, case series also serve as valuable intellectual background for future clinical or scientific work. For example, consider Dennis Burkitt\'s report on jaw tumors in African children, Alfred Blalock\'s initial efforts in cardiac surgery, or, more recently, Starzl and colleagues\' observations, in a small collection of patients, of donor leukocyte chimerism, whereby recipients acquire tolerance to foreign donor cells. In all three cases, the authors\' work led to powerful shifts in our understanding of the biology and treatment of disease \[[@pmed-0010013-b2],[@pmed-0010013-b3],[@pmed-0010013-b4]\]. All were case reports or case series---but under the current paradigm adopted by most journals and evidence-based databases, they would not be valued \[[@pmed-0010013-b5],[@pmed-0010013-b6],[@pmed-0010013-b7]\].[](#pmed-0010013-g001){ref-type="fig"} ::: {#pmed-0010013-g001 .fig} ::: {.caption} ###### Surgical research needs to move from case series to RCTs (Photo by Linda Bartlett, National Cancer Institute) ::: ![](pmed.0010013.g001) ::: Barriers to Surgical RCTs {#s3} ========================= There are many reasons why RCTs in surgical patients may be more difficult to perform than those in non-surgical patients. One of the most important---though least understood---is that the complexities of human disease in surgical patients makes them a more difficult group to study. Surgical patients are often heterogeneous in many more ways than non-surgical patients. So it would be inherently easier, for example, to study a new medication for generally healthy young adults with essential hypertension than a surgical technique for older patients with hepatic failure needing transplantation. In addition, while there may be value in studying patients from multiple centers, there may be important differences in the skill levels of different surgeons, either between centers or across the country. For example, the skill levels of surgeons in trials of carotid endarterectomy may be greater than those across the surgical community as a whole. This makes the applicability of some surgical RCTs to the wider community less certain than trials of medical therapies. So when it comes to surgical research, for both researchers and funding agencies, it is easier to grapple with a difficult, but ultimately soluble, basic science question than to face the uncertainty of clinical research. Investigators understand these implicit issues and trim their sails accordingly. Improving the Rigor of Research {#s4} =============================== Nonetheless, too much surgical work is conducted in the less rigorous format of the case series. What can and should be done to improve the rigor of surgical investigation? It would seem that improvements are required from within and beyond the surgical world. First, as Horner observed, and several eminent surgeons have since agreed, reforms must begin within the field itself \[[@pmed-0010013-b1],[@pmed-0010013-b5],[@pmed-0010013-b6],[@pmed-0010013-b7]\]. Both during surgical training and in the early years of faculty development, surgeons must obtain a thorough grounding in the principles of basic research and proper clinical investigation. Second, surgeons must establish firm and friendly relations with biostatisticians so that the latter may play a strong role in helping to develop adequately powered studies that can answer critical questions raised by new therapies and techniques. This is an especially acute need in an accelerating age of targeted therapies and disease biomarkers. Third, surgeons must re-engage in the clinical research enterprise and resume leadership roles in local and national clinical trials that involve surgical patients. In the United States, for example, an important step in this regard has been the establishment of the American College of Surgeons Oncology Group, which invites surgeons from all sectors, including private practice, to become active participants in well-designed, multi-institutional trials \[[@pmed-0010013-b5]\]. Similar efforts are needed on a global level. Finally, similar to the pressures faced by their colleagues elsewhere in academia, surgeon clinician-investigators must be nurtured, protected, and valued by their colleagues and medical administrators. The financial health of academic medical centers relies heavily on the generation of clinical revenue, which in many centers falls disproportionately on the shoulders of surgeons. New paradigms for revenue generation and funding of clinical research are needed. Funding for Surgical Research {#s5} ============================= Beyond the walls of the academic medical center, there also needs to be greater recognition of the value of scientifically sound surgical research and clinical investigation. However, the National Institutes of Health (NIH), the major source of biomedical funding in the United States, continues to convey a less welcoming attitude toward surgical research than toward other types of clinical or basic science\[[@pmed-0010013-b8],[@pmed-0010013-b9]\]. At the NIH, the principal instrument for performing peer review and making grant funding decisions is the study section, composed of about 10--20 members with expertise in a given field. There are few study sections devoted to surgically oriented clinical research and only two study sections (from among more than 100) in which surgeons make up even a reasonable minority of the committee members \[[@pmed-0010013-b8]\]. In comparison to those in other clinical departments, surgical grant proposals are less likely to be funded, and awards, when funded, are smaller \[[@pmed-0010013-b8]\]. "Funding agencies need to recognize the importance of the surgical endeavor to modern medicine." Surgical research is also impeded by processes affecting other types of research as well. The number of researchers under 35 years of age receiving a first RO1 grant, the main NIH mechanism for external funding, in any field, is below 4%. The average age of initial funding for US physicians is about 44 years, and shows a trend toward advancing age that has progressed significantly in the past two decades. Thus, the NIH appears to reward experience and proven results very heavily, which may stifle innovation and likely serves as an innate barrier for younger physician-investigators contemplating research careers \[[@pmed-0010013-b9]\]. To help correct for this worrisome trend, the NIH created the "K" award system---career development grants designed to help starting researchers gain the experience needed to compete for RO1 grants. However, nearly 40% of the clinicians who receive KO8 awards never apply for RO1 funding \[[@pmed-0010013-b10]\], which suggests that the overall support---both explicit and implicit---for clinical research at the institutional and funding levels is inadequate. Finally, outside the US, surgeons face similar, if not greater problems. This bodes poorly for countries where the cost of evaluating new therapies and technologies may be an unaffordable luxury. These challenges to the surgical research enterprise are therefore global issues and should merit the attention of surgeons, medical institutions, and funding agencies in all countries. The Future {#s6} ========== What can be done? On the national and international level, funding agencies need to recognize the importance of the surgical endeavor to modern medicine. Recently, in the US the NIH unveiled a "roadmap" (<http://nihroadmap.nih.gov>) designed to provide "new pathways to discovery." Clear, careful, scientific surgical investigation must be part of this roadmap, although it is not specifically mentioned. Outreach efforts to include surgeons in a variety of study sections should be made to ensure that important insights into the pathophysiology and treatment of disease, with which surgeons are concerned on a daily basis, are not overlooked. Additional efforts are needed to improve funding for clinical research, both for individuals at early stages of their careers and for multi-disciplinary clinical research and clinical trials. Locally, and individually, surgeons must join efforts to improve the clinical research enterprise by including training in clinical investigation at an early stage in medical school and during surgical residency training, fostering the careers of young surgeon-investigators through committed, protected time, participating in local and national clinical research groups, and recognizing that development as a clinical researcher takes time---many years in fact. These efforts may help ensure that surgical research is a vital part of the future of medicine and that it leads to the kind of high-quality work that shapes and remodels the face of medicine. To foster these efforts, surgeons must change and adapt to the currents of modern medical research. If this is successful, the case series will become the occasional rather than the common form of surgical communication. And surgeons, other clinicians, and, most importantly, basic scientists will be better able to take advantage of the new avenues of biomedical science opening before us. But the case series will always represent one important tool for early studies or uncommon conditions. It remains true that while the method one uses influences the answer one receives, it can be just as important to ask the right questions, which can be asked even in a series of one patient \[[@pmed-0010013-b11]\]. And surely that is the place one must begin. Citation: Weil RJ (2004) The future of surgical research. PLoS Med 1(1): e13. NIH : National Institutes of Health RCT : randomized controlled trial [^1]: Robert J. Weil is the associate director for basic research at the Brain Tumor Institute at the Cleveland Clinic Foundation, Cleveland, Ohio, United States of America. E-mail: <weilr@ccf.org> [^2]: Competing Interests: The author was formerly a member of the intramural research program of the National Institute of Neurological Disorders and Stroke at the National Institutes of Health.	PubMed Central	2024-06-05T03:55:48.584091	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523834/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e13", "authors": [ { "first": "Robert J", "last": "Weil" } ] }
PMC523835	Violent political conflict, and its impact, is again on the front pages---in Iraq, Afghanistan, and Sudan. While the situation in Darfur is now particularly urgent (see sidebar) \[[@pmed-0010014-b1],[@pmed-0010014-b2],[@pmed-0010014-b3],[@pmed-0010014-b4],[@pmed-0010014-b5]\], there are many other settings in which complex political emergencies are undermining health service provision and threatening human rights. Such emergencies have a direct impact on health (see [Table 1](#pmed-0010014-t001){ref-type="table"}). They also impair the functioning of health systems through, for example, destruction of infrastructure (such as clinics and vehicles), reduced access to medicines, death of health workers, and weakened national capacity for health policy-making \[[@pmed-0010014-b6]\]. ::: {#pmed-0010014-t001 .table-wrap} Table 1 ::: {.caption} ###### Examples of the Direct Impact of Conflict on Health ::: ![](pmed.0010014.t001) Adapted from \[[@pmed-0010014-b7]\], with permission from the World Health Organization ::: Such violent political conflicts stir us---the global health community---to discover our own humanity amidst the bloodshed. How best should we respond? Despite unique features in each setting, we must learn lessons from previous conflicts to help guide our response to current and future ones. There are six key lessons that emerge from studying health in conflict settings. Lessons from Conflict Settings {#s2} ============================== Violent conflict is driven by politics and economics \[[@pmed-0010014-b7]\]. Complex political emergencies (1) occur within and across state boundaries, (2) have political antecedents typically relating to competition for power and resources, (3) are protracted in duration, (4) express existing social, political, economic, and cultural structures and cleavages, and (5) are often characterised by one sector preying on other parts of the community \[[@pmed-0010014-b8]\]. Damage to health is not just a side effect but may be the objective for violent groups. In complex political emergencies, we can typically identify three groups: the winners, the 'conflict entrepreneurs' (who seek the perpetuation of conflict because they profit economically or politically), and the losers, whose lives and livelihoods are imperilled. Humanitarian and relief agencies increasingly recognise that belligerents may seek to control or manipulate the inflow of humanitarian and relief resources \[[@pmed-0010014-b9]\]. A political economy perspective helps identify those interests, which may impede the transition to peace \[[@pmed-0010014-b7]\]. ::: {.caption} ###### Sudan---Conflict and Health ::: The current crisis in Darfur reflects a devastatingly acute episode in the chronic internal conflict that has plagued Sudan since 1983. The cost of this conflict has been enormous: over 2 million lives lost, over 628,000 refugees from Sudan in neighbouring countries, and over 4 million people internally displaced \[[@pmed-0010014-b1]\]. In southern Sudan, the conflict has led to widespread ill health and has severely compromised the well-being of women and children. Indicators of immunisation, nutrition, primary school completion, and antenatal care are among the worst in the world. About 95,000 children under five years old died last year, most from preventable disease \[[@pmed-0010014-b2]\]. Statistics from UNICEF are chilling: 'A girl born in southern Sudan has a better chance of dying in pregnancy or childbirth than of completing primary school....One in nine women dies in pregnancy or childbirth but only one in a hundred girls completes primary school' \[[@pmed-0010014-b2]\]. Communities in Darfur face ongoing violence from militia supported by the government of Sudan. The fighting has resulted in large-scale destruction of villages, rape, and kidnapping. About 15,000--30,000 lives are estimated to have been lost from January 2003 to June 2004 \[[@pmed-0010014-b3]\]. Surveys by Médecins Sans Frontières found death rates of three to five per 10,000 people/day in Mornay and Zalinge villages (the emergency threshold level is set at one death per 10,000/day) \[[@pmed-0010014-b4]\]. Over 300,000 people are at risk if humanitarian access remains restricted. Of displaced Darfurians, 90% need shelter and latrines, and over half lack access to primary health care \[[@pmed-0010014-b3]\]. Food insecurity is widespread and is being used as a 'weapon of war' \[[@pmed-0010014-b5]\] resulting in widespread nutritional problems. Despite widespread concern, information gaps abound, and humanitarian agencies report having access to only a fraction of those most affected. Yet, genocide is taking place in real time. Appreciating context is crucial. The nature of the conflict---its background, history, and the different forms of violence involved---will greatly influence health outcomes. Most conflicts are today intra-national rather than international \[[@pmed-0010014-b10]\]. Internal conflicts affect populations through forced migration, violence, and human rights abuses including torture, disappearances, and rape. The forms of violence and types of health damage relate to the phase of the conflict, the sophistication of weapons used, the degree of involvement of regular military forces, the extent of terrorism employed, and the extent to which genocide is intended. Ongoing insecurity and instability may be present even after the ostensible end to the conflict, as in latter-day Afghanistan and Iraq. Challenges to governance, to service delivery, and to the reestablishment of livelihoods may persist for years. A 2003 survey in Iraq found that despite the brief duration of the war and the intent to spare hospitals and clinics from direct attack, many people suffered in the post-war period, primarily as a result of disruption to civil order \[[@pmed-0010014-b11]\]. Recent reports highlight the difficulties of re-establishing the health system in Iraq---partly because of a failure to appreciate the cultural and health services context \[[@pmed-0010014-b12]\]. Better care can save lives. Emergency relief efforts are increasingly based upon empirical evidence, and priority health issues are much more effectively addressed than previously. Emphasis is typically placed upon disease surveillance, immunisation, control of infectious diseases, reproductive health, water and sanitation, shelter, and nutrition \[[@pmed-0010014-b13]\]. Mental health, sexually transmitted infections, and HIV have recently attracted additional attention. Standards have improved, can be further improved, and warrant widespread dissemination and application. The more-established humanitarian agencies have accepted that their relief efforts must be as evidence-based as possible. This principle should also apply to the post-conflict period, during which the health of affected communities continues to suffer \[[@pmed-0010014-b14]\].[](#pmed-0010014-g001){ref-type="fig"} ::: {#pmed-0010014-g001 .fig} ::: {.caption} ###### About 70% of structures were destroyed in Dili, East Timor, in the violence wrought by Indonesian militia after the referendum in 1999 (Photo: Anthony Zwi.) ::: ![](pmed.0010014.g001) ::: We need enhanced accountability for humanitarian action. Despite a developing evidence base for health-related humanitarian action, evaluations of humanitarian activities have found ongoing problems. These include poor standards of delivery, duplication of efforts by different agencies, lack of coordination, and failing to learn from prior experience. The Sphere Project has advocated minimum standards for the delivery of humanitarian assistance, and has established a "Humanitarian Charter" (<http://www.sphereproject.org>). The project\'s objectives and achievements have been to improve the quality of humanitarian action and promote a movement concerned with the rights and dignity of those caught up in war and disaster \[[@pmed-0010014-b15],[@pmed-0010014-b16]\]. The Active Learning Network for Accountability and Performance in Humanitarian Action (<http://www.alnap.org>) seeks to ensure that lessons are learned, distilled, and disseminated. At a meeting in Stockholm in June 2003, key international donors committed themselves to 'good humanitarian donorship', which recognises the importance of promoting standards in humanitarian action \[[@pmed-0010014-b17]\]. However, recent sober reflection suggests that donors and humanitarian agencies could do better: 'An ailing humanitarian enterprise is labouring under pressures from the external environment over which it has little control, while struggling with issues internal to its own function for which it should take greater responsibility' \[[@pmed-0010014-b18]\]. Militarization of humanitarian efforts is problematic. Multinational military forces have played a major part in recent conflicts in Kosovo, East Timor, Sierra Leone, Iraq, and Afghanistan. The military has become increasingly involved not only in waging war but also in seeking to win the peace; it is increasingly active in delivering emergency relief. It not only provides services---sometimes necessary to deliver needed relief---but also seeks to 'win hearts and minds' while operating within structures responsive to military and foreign policy directives. The result has seen a blurring of the separation between military and humanitarian efforts \[[@pmed-0010014-b19]\]. This can make humanitarian agencies a target---recent examples include the bombing of United Nations headquarters and the International Committee of the Red Cross in Iraq and the recent, reluctant withdrawal of Médecins Sans Frontières from Afghanistan following the murder of five aid workers \[[@pmed-0010014-b20]\]. Emerging evidence and good practice in civil-military cooperation highlights the importance of (1) promoting needs-based assistance free of discrimination, (2) civilian-military distinction in humanitarian action, (3) independence of humanitarian organisations from political pressures and interference, and (4) the security of humanitarian personnel \[[@pmed-0010014-b19]\]. The transition from emergency relief to development is poorly managed. The objectives of humanitarian relief activity (saving lives and livelihoods) differ from those of development (building sustainable systems, promoting equity, building systems of governance, and eradicating poverty). In each phase there are different actors, strategies, and approaches. The increasing politicisation of humanitarian intervention \[[@pmed-0010014-b21],[@pmed-0010014-b22]\] brings threats and dangers, undermining key humanitarian principles. The balance between relief and development will vary over time and place; getting the balance right and adequately resourcing the transition warrants careful research, documentation, reflection, and the commitment of appropriate longer-term funding. What Gaps Remain in Our Knowledge? {#s3} ================================== Despite the knowledge we have gained on responding to violent political conflict, many important gaps remain.[](#pmed-0010014-g002){ref-type="fig"} ::: {#pmed-0010014-g002 .fig} ::: {.caption} ###### In southern Sudan, the conflict has affected the well-being of women and girls (Illustration: Margaret Shear, Public Library of Science.) ::: ![](pmed.0010014.g002) ::: We still do not hear the voices of those most affected or of the service providers seeking to assist. The reality of people\'s experiences is inadequately appreciated \[[@pmed-0010014-b23]\]; whatever we learn of their fears, challenges, and suffering is typically represented and reported through sanitised language and media. The language used dehumanises 'the enemy' and blunts our senses to the reality of atrocity and to the negative effects of our own countries' interventions. Within the health sector, ensuring that we hear the voices of service providers and carers will help bring home the reality of system disruption, destruction, and damage and will simultaneously document the mechanisms and potential for effective responses. The new communication technologies provide immense opportunity to ensure that experience is placed in the public domain from where lessons can be drawn and better practice promoted. We know little about how communities and systems survive adversity. In most settings, the inherent ability and ingenuity of people and systems allows them to withstand instability and insecurity. Health personnel and health systems could play a valuable role in these fragile settings---assisting individuals, communities, and systems to further develop their coping strategies, adaptations, and responses. But whether health systems do so and how is unclear. Failing to support and maintain these systems may result in much greater challenges when we seek at a later stage to resuscitate them. We also know relatively little about whether the health sector can indeed make a special contribution to building the peace. While it has been forcefully argued that the health sector is uniquely placed to play a role in peace building \[[@pmed-0010014-b24]\], the evidence for this remains limited \[[@pmed-0010014-b25]\]. We know little about how health workers see and respond to these challenging roles. The health sector could play a role in demonstrating the values and priorities of government, reflecting the relationship between those with and without resources, and the relationship between those who do and do not have protection. In the aftermath of major periods of violence, the health sector could also help to ensure that the structural inequities that preceded the violence and may have contributed to it, are not reinforced and the same injustices not recreated. But, engagement around health is not always positive: the health system is open to abuse and has been abused by repressive systems. From Learning Lessons to Sound Policy {#s4} ===================================== Perhaps the most important gap of all is between observing lessons and putting them into practice. We urgently need to transform evidence and experience into sound policy. We need more sophisticated policy analyses, more sensitive policy-making, and more relevant research. Policy in these difficult areas will never be entirely evidence-based---often it will at best be 'evidence-informed'. Our objective must be to promote organisations and systems that are able to reflect on experience, work with partners to critically analyse and learn, and thereby formulate better responses. Violent political conflict will continue to challenge the global health community. International policy-makers and funders must support more extensive documentation and reflection: the building blocks of better practice. Natalie Grove assisted in identifying and summarising key source documents cited in this contribution. Citation: Zwi AB (2004) How should the health community respond to violent political conflict? PLoS Med 1(1): e14. [^1]: Anthony B. Zwi is professor and head of the School of Public Health and Community Medicine at The University of New South Wales, Sydney, Australia. E-mail: <a.zwi@unsw.edu.au> [^2]: Competing Interests: The author is chief investigator of a project seeking to explore the links between health and preventing violence, and health and building peace, in five countries in the Asia-Pacific region. This project is funded by AusAID, the Australian development assistance agency.	PubMed Central	2024-06-05T03:55:48.585903	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523835/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e14", "authors": [ { "first": "Anthony B", "last": "Zwi" } ] }
PMC523836	Case Discussion {#s1} =============== Bupropion for Smoking Cessation {#s1a} ------------------------------- Originally developed as an antidepressant, bupropion has more recently been licensed in many countries as an aid to smoking cessation. It came onto the Mauritian market as a smoking cessation aid in November 2003. The British National Formulary ([www.bnf.org](www.bnf.org)) recommends starting the drug one to two weeks before the target smoking stop date, initially at a dose of 150 mg daily for 6 d and then 150 mg twice daily. The maximum period of treatment is 7--9 wk; treatment should be discontinued if abstinence is not achieved by 7 wk. The efficacy of bupropion as an aid to smoking cessation has been shown in randomized, double-blind, placebo-controlled trials \[[@pmed-0010015-b01],[@pmed-0010015-b02]\]. But there have also been reports of death, seizure, serum sickness, generalized acute urticaria, myocardial infarction, and psychosis in people taking the drug \[[@pmed-0010015-b03],[@pmed-0010015-b04],[@pmed-0010015-b05],[@pmed-0010015-b06],[@pmed-0010015-b07]\]. Our report is of a woman who had a generalized seizure on the sixth day of treatment with bupropion; she had no other risk factors for seizures. Contraindications to Bupropion {#s1b} ------------------------------ To reduce the risk of seizures, the drug should not be given to patients with a current seizure disorder or any history of seizures, with a current or previous diagnosis of bulimia or anorexia nervosa, with a known central nervous system tumour, or to those experiencing abrupt withdrawal from alcohol or benzodiazepines \[[@pmed-0010015-b08],[@pmed-0010015-b09]\]. The United Kingdom Medicines Control Agency states that bupropion must not be prescribed in patients with other risk factors for seizures, unless there is a compelling clinical justification for which the potential medical benefit of smoking cessation outweighs the potential increased risk of seizure \[[@pmed-0010015-b08]\]. Predisposing risk factors for seizure include the following: concomitant use of medications known to lower seizure threshold (including antipsychotics, antidepressants, antimalarials, tramadol, theophylline, systemic steroids, quinolones, and sedating antihistamines), history of head trauma, diabetes treated with hypoglycemics or insulin, history of alcohol abuse, and use of stimulants or anorectic products \[[@pmed-0010015-b08]\]. The Seizure Risk {#s1c} ---------------- Bupropion is associated with a dose-related risk of seizure. The Medicines Control Agency states that the incidence of seizures is one in 1,000 based on doses up to the maximum recommended daily dose of 300 mg per day \[[@pmed-0010015-b08]\]. The seizure risk may be reduced by taking no more than 150 mg (1 pill) at a time and, if taking two daily doses of 150 mg each (two pills per day), ensuring that doses are taken at least 8 h apart (see [www.bnf.org](www.bnf.org)). Up to 24 July 2002, in the UK there were 184 reports of seizures suspected as being associated with the use of bupropion \[[@pmed-0010015-b08]\]. In about half of the reports, patients had a past history of seizures and/or risk factors for their occurrence. The Presenting Case {#s2} ------------------- In the case we have presented, the patient had no current or previous history of seizure disorders, bulimia, or anorexia nervosa. She was on the sixth day of treatment with bupropion, taking 300 mg per day in two separate doses. She had no other predisposing risk factors for seizure. In particular she was taking no prescription medications, or over-the-counter medications (such as those containing ephedrine products) that are known to lower the seizure threshold. While systemic steroids can lower the seizure threshold, systemic absorption of the nasal steroid drops the patient was taking is an extremely unlikely cause for her seizure (she was not taking a high dose of high-strength drops). Her weight (48 kg) may have been an important factor, since bupropion reaches higher plasma levels in smaller individuals---indeed clinical trials of the drug have generally excluded patients weighing under 100 lb (about 45 kg) \[[@pmed-0010015-b02]\]. ::: {.caption} ###### Learning Points ::: This case report is a useful reminder to clinicians that bupropion is associated with a dose-dependent risk of seizures.In about half of the reports of seizure associated with bupropion, there was a past history of seizures and/or risk factors for their occurrence.It is extremely important to adequately assess seizure risk before prescribing bupropion.Patients should be made aware of the possible adverse effects of the drug. Citation: Lan Cheong Wah MF, Lan Cheong Wah LSH (2004) Generalized seizure in a Mauritian woman taking bupropion. PLoS Med 1(1): e15. [^1]: Competing Interests: The authors have declared that no competing interests exist. [^2]: Author Contributions: Both authors contributed equally to preparing the case report.	PubMed Central	2024-06-05T03:55:48.587394	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523836/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e15", "authors": [ { "first": "Marie France Lan Cheong", "last": "Wah" }, { "first": "Lan Sem Hing Lan Cheong", "last": "Wah" } ] }
PMC523837	Introduction {#s1} ============ Malaria claims one million lives annually, with more than 90% of these being those of children in sub-Saharan Africa \[[@pmed-0010018-b01]\]. Most deaths of hospitalised children occur in the first 24 h after admission. Even modest improvements in management during this time may improve survival \[[@pmed-0010018-b02]\]. There is considerable disagreement about the degree to which children with severe malaria become hypovolemic. In east African studies, clinical signs of severe malaria (such as tachycardia, prolonged capillary refill times, and decreased urine volume) have been interpreted as evidence for volume depletion \[[@pmed-0010018-b02],[@pmed-0010018-b03],[@pmed-0010018-b04],[@pmed-0010018-b05],[@pmed-0010018-b06]\]. However, determining fluid compartment volumes is the first and most critical step in optimising fluid replacement therapy for children with malaria because clinical assessment of fluid status is difficult and imprecise \[[@pmed-0010018-b07]\]. Our study was designed to measure total body water volume (TBW) and extracellular water volume (ECW) using nonradioactive tracer dilution techniques and to derive intracellular water volume (ICW). Bromide distributes in the extracellular space so that concentrations measured 2--4 h after administration safely and reliably estimate ECW. Heavy water (^2^H~2~O) space represents TBW. ICW is calculated by subtraction of the ECW from the TBW. Tracer dilution methods are expensive and invasive and cannot be repeated at short intervals. Therefore, we simultaneously calibrated a noninvasive technique of bioelectrical impedance analysis (BIA) \[[@pmed-0010018-b08]\] to estimate the fluid volumes. BIA measures the opposition (impedance) of the body to the flow of a small alternating current between electrodes placed on the hand and the foot, and then estimates TBW and ECW using regression equations derived by calibration against 'gold standard\' tracer measurements of fluid volumes. We hypothesised that volume changes in fluid compartments would reflect disease severity in malaria and that these changes would be related to established markers of disease severity \[[@pmed-0010018-b09]\]. We also calibrated BIA assessments in children with moderate and severe malaria with direct measurements of volume of TBW and ECW. Methods {#s2} ======= The study was conducted at the Albert Schweitzer Hospital, Lambaréné, Gabon, and Centre Hospitalier de Libreville, Gabon. It was approved by the ethics committees of the International Foundation of the Albert Schweitzer Hospital, the Gabonese Ministry of Health, and the University of Tübingen. Children (aged 1 to 10 y, inclusive) admitted with suspected severe or moderate malaria were referred to the study team, who assessed them within 15 min, and the children were admitted to the study once informed consent had been obtained from the parents. Malaria was defined as the presence of asexual forms of Plasmodium falciparum in thick or thin blood films. Severe malaria was malaria with one or more of the following features: blood lactate ≥ 5 mmol/l, blood glucose ≤ 2.2 mmol/l, Blantyre coma score ≤ 2, or repeated, observed seizures \[[@pmed-0010018-b02]\]. Moderate malaria was malaria without any of the features of severe malaria but with a requirement for parenteral treatment because of one or more of the following: a history of frequent (\> 2) and recent vomiting (within 12 h), drowsiness, obtundation, or prostration \[[@pmed-0010018-b02]\]. Alternative diagnoses were excluded clinically. Assessment and Management {#s2a} ------------------------- On admission children were weighed (undressed) with pediatric scales accurate to within 100 g (Seca, Birmingham, United Kingdom). A history was taken from parents, and the child was examined with particular attention to signs of dehydration, including: capillary refill time, skin turgor, sunken eyes, dry mucous membranes, and absence of tears. Vital signs, blood glucose and lactate concentrations, and hematocrit and parasitaemia were measured every 4 h for the first 24 h and then every 6 h until recovery. If peripheral venous access was impracticable, then a femoral central venous catheter was inserted, and the central venous pressure was measured every 4 h with a manometer zeroed at the midaxillary line. Children were managed in a standard manner as previously described \[[@pmed-0010018-b10],[@pmed-0010018-b11],[@pmed-0010018-b12]\]. All received intravenous quinine (20 mg/kg salt intravenously as a loading dose given over 4 h, then 10 mg/kg intravenously every 12 h until able to take oral medication) (Quinimax, Sanofi Synthelabo, Paris, France). Hypoglycaemia (blood glucose ≤ 2.2 mmol/l) was treated with 25% glucose (2 ml/kg). Convulsions were treated with diazepam (0.3 mg/kg intravenously or 0.5 mg/kg intrarectally) (Roche, Basel, Switzerland), and repeated convulsions were treated with phenobarbital (7.5 mg/kg intramuscularly). In addition to a 5% or 10% dextrose infusion (at least 3 mg/kg/min, i.e., 1.6 or 3.2 ml/kg/h), physicians were free to give any fluid replacement regimen as clinically indicated, including boluses of saline or blood. A strict fluid input/output chart was kept for each child. A blood transfusion of 20 ml/kg of cross-matched whole blood tested for blood-borne pathogens was given over 4 h if the hematocrit fell below 15%. Measurement of TBW and ECW {#s2b} -------------------------- A sterile standard dosing solution was prepared by adding 119 ml of ^2^H~2~O per litre of 2.315% sodium bromide solution. A baseline sample of blood (3 ml) was drawn, and the plasma frozen at −70 °C for bromide and ^2^H~2~O assays. At the start of the study, 2.8 ml/kg of the dosing solution was administered intravenously over 20 min. Four hours after injection a second blood sample was drawn (1.5 ml) for the determination of blood bromide and ^2^H~2~O concentrations. Parents were asked to return with their children 28 d after admission; children were examined, and measurement of TBW and ECW repeated. ^2^H enrichment was measured in duplicate by isotope ratio mass spectrometry, using a Sira 10 instrument (Micromass, Cheshire, United Kingdom) as described \[[@pmed-0010018-b13]\]. The precision of the TBW determination was estimated at 0.3% of the value obtained. Batch analysis of bromide enrichment in plasma was performed by high-performance anion-exchange liquid chromatography as described \[[@pmed-0010018-b14]\]. The intra-assay coefficient of variation for bromide was better than 1.5%. BIA was performed using a SEAC SFB3 multifrequency bioimpedance meter (Impedimed, Brisbane, Australia). An alternating electrical current of 200 μAmp was applied between 2 Ag/AgCl electrodes at the right hand and right foot. Whole body and segmental impedance were measured by rotating the sensing Ag/AgCl electrodes between four sites on the ankles or wrists as described \[[@pmed-0010018-b13]\]. Each set of measurements was taken at 496 frequencies between 4 kHz and 1012 kHz. Measurements were taken at 0 h, 4 h, 12 h, 24 h, and discharge. The process took about 2 min to perform and was simpler than obtaining an electrocardiogram. Data were analysed with software (Bioimp, version 1.1.0, Impedimed) that uses nonlinear regression to fit measured data to semicircular Cole-Cole plots \[[@pmed-0010018-b15]\] of reactance against resistance. Resistance and reactance values, obtained from the Cole-Cole plot for specific frequencies (0, 4, 50, and 100; characteristic and infinite kHz) were used in further analysis \[[@pmed-0010018-b13]\]. Plasma electrolytes were measured (Beckman Coulter, Allendale, New Jersey, United States), and osmolality was calculated using freezing point depression on a Micro Osmometer (Vitech Scientific, West Sussex, United Kingdom). Osmol gap was calculated as described \[[@pmed-0010018-b16]\]: where OG is osmol gap, MO is measured osmolality, \[Na^+^\] is plasma sodium concentration (millimoles/litre), \[glucose\] is plasma glucose concentration (millimoles/litre), \[urea\] is plasma urea concentration (millimoles/litre), 1.86 is a correction factor as sodium chloride is only 93% dissociated, and 0.93 is the assumed proportion of water in plasma. Statistical Methods {#s2c} ------------------- Statistical analyses were carried out using Stata Statistical Software (Releases 6.0--8.0, College Station, Texas, United States). After checking distributions with the Shapiro-Wilks W test, and transforming data logarithmically if necessary, we analysed normally distributed data by two-tailed Student\'s t test and nonparametric data with the Wilcoxon sign-rank test. Proportions were compared with Fisher\'s exact test, and correlations were assessed by linear regression analysis of Pearson or Spearman. Predictive values for TBW and ECW from impedance measurements were obtained by multivariate analysis in a backward elimination process with p \< 0.05 for entry and p \< 0.10 for exit, confirmed by forward selection. This analysis was based on the first pair of BIA and isotope dilution measurements after admission. In a second step, whole body BIA models were replaced with segmental BIA models as previously described \[[@pmed-0010018-b13]\] and compared with whole body estimates. Errors in the BIA estimate and isotope dilution methods were compared. TBW estimates were corrected for fluid input and output during the period of measurement. Sample size was calculated from previously published values \[[@pmed-0010018-b17],[@pmed-0010018-b18]\] assuming a mean (SD) for convalescent TBW of 586 (30) ml/kg; we wished to detect a 5% difference in TBW between severe and moderate cases with a power of 90%. Results {#s3} ======= Between October 1999 and March 2000, 205 children who were judged ill enough to be hospitalised were referred to the study team. One hundred and thirty-two children had malaria, 20 with severe and 35 with moderate disease. Ten children with moderate and two with severe malaria were ineligible for study because of their age, and for seven children (one with severe malaria) consent could not be obtained. One child died before inclusion into the study, leaving nineteen children with moderate and sixteen with severe malaria admitted to this study. The median (interquartile range \[IQR\]) time from admission to administration of tracers was 54 (37--84) min, during which complications such as convulsions or hypoglycaemia were treated. The baseline characteristics of children are given in [Table 1](#pmed-0010018-t001){ref-type="table"}. Those with severe malaria had significantly higher pulse rates, mean arterial pressure and blood lactate concentrations, and a longer capillary refill time (p \< 0.001), compared to children with moderate malaria. Capillary refill time and blood lactate concentrations were correlated with each other (adjusted r^2^ \[adj r^2^\] = 0.25, p = 0.031). ::: {#pmed-0010018-t001 .table-wrap} Table 1 ::: {.caption} ###### Baseline Characteristics of Children with Malaria ::: ![](pmed.0010018.t001) Mean (SD) / median (IQR) shown ^a^ Parasitaemia values are geometric mean (range) ::: The volume of fluid (including blood) given in the first 4 h of the study was similar in children with severe (median \[IQR, range\] of 3.7 \[2.2--5.8, 2.1--11.4\] ml/kg/h) and moderate (median \[IQR, range\] of 3.3 \[2.2--4.2, 1.2--15.1\] ml/kg/h) malaria (p \> 0.5). There was rapid correction of vital signs in all children ([Figure 1](#pmed-0010018-g001){ref-type="fig"}), and vital signs were similar in both study groups by 8 h. There were two deaths (4.5 and 6 h after admission), and two children with severe malaria had major persistent neurological deficits (4/16 \[25%\] with adverse outcomes). Central venous pressure (CVP) measurements were obtained in six children with severe malaria. The median (IQR) CVP on admission was +6.5 (3--7.5) cm H~2~O with no significant rise in the first 24 h. ::: {#pmed-0010018-g001 .fig} Figure 1 ::: {.caption} ###### Vital Signs of Children during the First 24 h after Admission Mean and 95% confidence interval shown. Red circles, severe malaria; blue triangles, moderate malaria. (A) Pulse (per minute), (B) mean arterial pressure (millimetres Hg), (C) respiratory rate (per minute), and (D) blood lactate concentration (millimoles/litre). ::: ![](pmed.0010018.g001) ::: Fluid Volumes {#s3a} ------------- Volume determinations using isotope dilution were available at baseline in all but one of the children with severe malaria. The values for the TBW, ECW, and ICW are given in [Table 2](#pmed-0010018-t002){ref-type="table"}. The TBW was significantly lower at admission compared with day 28 for the severe cases (p = 0.028) but not for moderate cases (p = 0.109). The mean (SD) TBW was lower in severe than moderate malaria at admission: 524 (44) ml/kg versus 555 (50) ml/kg (p = 0.052). The mean (SD) change in TBW between admission and follow up was 48 (42) ml/kg and 12 (37) ml/kg for severe and moderate cases, respectively. Individual data from the children with severe malaria are shown in [Table 3](#pmed-0010018-t003){ref-type="table"}. ::: {#pmed-0010018-t002 .table-wrap} Table 2 ::: {.caption} ###### Volumes Determined by Tracer Dilution and Electrolyte Measurements ::: ![](pmed.0010018.t002) Mean (SD) / median (IQR) shown ^a^ Change in volume as measured by isotope dilution between admission and day 28 ^b^ Measurements of ECW and ICW were available at follow up in three cases with severe and ten with moderate malaria (discrepancy due to difficulties obtaining adequate sample volumes) ::: ::: {#pmed-0010018-t003 .table-wrap} Table 3 ::: {.caption} ###### Individual Details of Severely Ill Children ::: ![](pmed.0010018.t003) ^a^ Missing values because discharge weights unobtainable ::: Bromide space (ECW) measurements were available in all but two children with severe malaria and in all but one of the children with moderate malaria. At baseline, the mean (SD) ECW was significantly lower in children with moderate than in those with severe malaria (p = 0.045). Admission ICW measurements were significantly lower in children with severe malaria than in those with moderate malaria (p \< 0.001). Bioelectrical Impedance Analysis {#s3b} -------------------------------- As predicted by theory \[[@pmed-0010018-b08]\], there was a strong correlation between height^2^ divided by impedance at 50 kHz (H ^2^/Z ~50~) and measured TBW. The 'best fit\' regression equation to predict TBW from BIA was derived using the variables age, weight, and H ^2^/Z ~50~ (standard error of the estimate \[SEE\] = 0.435, adj r^2^ = 0.975): where A is age (months), H is height (centimetres), W is weight (kilograms), and Z ~50~ is impedance at 50 KHz frequency. Disease severity and gender did not contribute significantly to this model, allowing data from admission to be pooled in this prediction equation. Repeating the analysis using impedances measured at other frequencies (4--1012 kHz) did not show a clear advantage, so all data for TBW prediction are given for measurement at 50 kHz. By contrast, ECW was not significantly associated with age or weight, and in agreement with previous studies in babies \[[@pmed-0010018-b19]\] H ^2^/R ~0~ emerged as the strongest predictive term (SEE = 0.584, adj r^2^ = 0.753): where H is height (centimetres), and R~0~ is resistance at zero frequency. Again, disease severity and gender did not contribute significantly to this model. Prediction equations based on segmental BIA data for ECW and TBW were inferior to whole body BIA models and are therefore not shown here. Errors in BIA estimates of TBW and ECW compared to values measured by isotope dilution are displayed with 95% limits of agreement in [Figure 2](#pmed-0010018-g002){ref-type="fig"}A and [2](#pmed-0010018-g002){ref-type="fig"}B, respectively. ::: {#pmed-0010018-g002 .fig} Figure 2 ::: {.caption} ###### Plots of TBW and ECW Estimates from Isotope Dilution and BIA Calculation with Measured Values Filled circles, admission value; open circles, day 28 follow up value; dotted lines, 95% confidence intervals for values. (A) TBW and (B) ECW. D~2~O, heavy water. ::: ![](pmed.0010018.g002) ::: Fluid Volumes from BIA {#s3c} ---------------------- Fluid volume estimates were available from BIA in 14 children with moderate and 11 with severe malaria on admission and 16 with moderate and 12 with severe malaria at discharge. Values for the TBW, ECW, and ICW are shown in [Figure 3](#pmed-0010018-g003){ref-type="fig"}. ::: {#pmed-0010018-g003 .fig} Figure 3 ::: {.caption} ###### Body Fluid Compartment Volumes Derived from BIA on Admission and Discharge Red circles, severe malaria; blue triangles, moderate malaria. (A) TBW (litres/kilogram), (B) ECW (litres/kilogram), and (C) ICW (litres/kilogram). ::: ![](pmed.0010018.g003) ::: BIA-determined TBW was significantly lower in those with severe compared with moderate malaria (mean \[SD\] of 539 \[[@pmed-0010018-b32]\] versus 562 \[[@pmed-0010018-b30]\] ml/kg, p = 0.034). TBW increased from admission to discharge in children with severe malaria by 37 (33) ml/kg or 6.7% (6.0%) (paired t test, p = 0.012), but there were no significant changes in TBW for children with moderate malaria between admission and discharge. ECW did not differ between the severe and moderate groups on admission and did not change significantly between admission and discharge. Serial measurements of ECW did not show any rise during the first 4 d after admission, at discharge, or at day 28 in children with severe or moderate malaria (data not shown). ICW at admission was significantly lower in the severe malaria group than in the moderate malaria group, with a mean (SD) of 293 (17) ml/kg versus 325 (28) ml/kg (p = 0.002). ICW remained unchanged between admission and discharge in children with moderate malaria. In the severe malaria group mean (SD) of ICW at admission was 40 (22) ml/kg, or 11.7% (11.0%) lower at admission than at discharge (p = 0.002). There were no differences in the fluid volume measurements of the four children with adverse and those with good outcomes. There were no relationships found between TBW, ECW, or ICW, or changes in TBW, ECW, and ICW between baseline and discharge, and any of the following markers of severity in malaria: blood lactate concentration, coma score, plasma creatinine concentration, peripheral parasitaemia, blood glucose, coma recovery time, time to walk, time to eat, time to drink, length of hospital stay, or having a history of diarrhoea or vomiting. Admission weight relative to the weight at discharge showed a mean (SD)/median (IQR, range) percentage deficit of −4.3% (4.4%)/−4.0% (−1.3% to −6.1%, −0.1% to −11.0%) for the children with severe malaria on admission (p = 0.002). For those with moderate malaria the deficit was not significant: −0.3% (4.6%) /−0.9% (1.6% to −1.5%, 7.0% to −8.1%) (p = 0.41). As predicted, the deficits in weight and TBW were correlated (adj r^2^ = 0.67, p \< 0.001). Electrolyte Measurements {#s3d} ------------------------ The concentrations of plasma electrolytes are shown in [Table 2](#pmed-0010018-t002){ref-type="table"}. The mean (SD) concentrations of sodium were significantly lower for the children with severe malaria than for those with moderate malaria (p = 0.036). Plasma potassium concentration and osmolality were higher in children with severe than in those with moderate malaria (p = 0.039 and p = 0.021, respectively). Plasma urea and creatinine were significantly higher in those with severe malaria. Osmol gap was significantly higher in children with severe than in those with moderate malaria (p \< 0.001), with 15/16 children in the severe group having an osmol gap greater than 8.2 mOsm (considered high in United States children \[[@pmed-0010018-b20]\]). Discussion {#s4} ========== We have shown that severe childhood malaria is associated with mild dehydration in most cases, with a mean (SD) depletion of TBW of 37 (33) ml/kg, or 6.7% (6.0%). Only 3/16 children (19%) in our study had moderate volume depletion (\> 60--90 ml/kg), and none were severely dehydrated (\> 100 ml/kg) \[[@pmed-0010018-b21]\]. Moderate malaria was not associated with any significant changes in TBW. Consistent with a lower TBW in severe disease, ICW was depleted in children with severe malaria by a mean (SD) of 40 (22) ml/kg, an 11.7% (11.0%) difference. However, we found no relationship between TBW, ECW, and ICW and clinical and laboratory markers of disease severity, in particular the two most important prognostic indicators of fatal outcome: hyperlactataemia and Blantyre coma score \[[@pmed-0010018-b09]\]. Our findings suggest that the degree of dehydration in children with severe malaria is unlikely to be a primary pathological process in the evolution of the disease. These findings are also consistent with our previous suggestion that hyperlactataemia arises from tissue hypoxia resulting from microvascular obstruction by infected erythrocytes \[[@pmed-0010018-b12],[@pmed-0010018-b22]\] rather than gross hypovolaemia. ECW did not change significantly during hospitalisation. Furthermore, our fluid replacement regimen (median \[IQR\] 3.7 \[2.2--5.8\] ml/kg/h) normalised vital signs and blood lactate within 8 to 12 h. Our estimates of fluid compartment volumes in children after recovery from malaria are entirely consistent with previous work in children (range: TBW, 540--640 ml/kg; ECW, 250--320 ml/kg; ICW, 260--340 ml/kg) \[[@pmed-0010018-b17]\], although our study is the first that we know of to examine fluid status in childhood malaria. Studies in adults with uncomplicated or moderate malaria have given conflicting results \[[@pmed-0010018-b23],[@pmed-0010018-b24],[@pmed-0010018-b25],[@pmed-0010018-b26],[@pmed-0010018-b27]\]. In sepsis, ECW increases by up to 50% of TBW because capillary permeability increases by up to 300% of normal. There are no significant increases in ECW in children with severe malaria, confirming earlier studies that indicate that sepsis and malaria syndromes result from different pathophysiological processes \[[@pmed-0010018-b09],[@pmed-0010018-b28]\]. Studies on fluorescein angiography in children and adults also confirm there is no increased capillary permeability in severe malaria \[[@pmed-0010018-b23],[@pmed-0010018-b29]\]. Furthermore, CVP measurements in a subgroup (6/16) of children were not low, and did not change significantly after 24 h of intravenous fluid replacement. Taken together, these findings do not suggest that volume depletion or increased capillary permeability are important to the pathophysiology of malaria in our population. Hyponatraemia \[[@pmed-0010018-b30]\] has been attributed to high and possibly inappropriate arginine vasopressin secretion \[[@pmed-0010018-b31]\] in severe malaria. Our findings (high osmol gap, low ICW, and normal ECW) are more in keeping with sick cell syndrome than with inappropriate arginine vasopressin secretion \[[@pmed-0010018-b32],[@pmed-0010018-b33]\]. To conclude that arginine vasopressin is inappropriately elevated, renal function must be normal and volume depletion excluded. No severely ill child in this study fulfilled these criteria. What are the implications of our findings for optimal fluid replacement therapy in malaria? We cannot answer precisely on the basis of measuring fluid compartment volumes because regimens are sometimes designed not only to correct existing fluid deficits and to provide maintenance requirements, but also to rehydrate more vigorously to maintain circulating volume. Such approaches are advocated by others for different populations of children with severe malaria, for example, in a series of studies published from Kilifi, Kenya \[[@pmed-0010018-b04],[@pmed-0010018-b05],[@pmed-0010018-b34]\]. However, estimates of fluid requirements for children with severe malaria have been based upon indirect measurements (such as monitoring vital signs and degree of acidosis) that are potentially misleading because they do not relate to the degree of fluid loss that we measured. Furthermore, adequately powered controlled studies aimed at defining appropriate fluid regimens for severe malaria are lacking, but should take into account our findings as well as the BIA methodology that we have now calibrated to measure fluid compartment volumes in malaria. Indeed, a fluid (0.9% saline) replacement rate of up to 20 ml/kg in 1 h is a considerably faster rate than we can advocate on the basis of our findings. In any case, BIA can now be used (equations [2](#pmed-0010018-e002){ref-type="disp-formula"} and [3](#pmed-0010018-e003){ref-type="disp-formula"}) to measure ECW and TBW noninvasively to guide treatment in patients with severe malaria. There are risks to over-vigorous fluid administration just as there are with inadequate fluid replacement, particularly in hospitals where measuring plasma electrolyte concentrations and providing assisted ventilation are difficult. These risks include pulmonary \[[@pmed-0010018-b03]\] and cerebral oedema \[[@pmed-0010018-b06]\] (which occur in adults and children, respectively) and dangerously rapid changes in plasma electrolyte concentrations. Children with severe malaria have a low ICW and are at risk of hypokalaemia if ICW is restored rapidly (\< 4 h), particularly when there is relative hyperinsulinaemia due to quinine administration \[[@pmed-0010018-b12]\]. A correlation between capillary refill time and blood lactate concentrations but none with fluid volume status suggests that prolongation of capillary refill time may be due to the common underlying process of microvascular obstruction. Our findings do not support the widespread use of aggressive fluid volume replacement in children with severe malaria. Clearly, volume depletion indicated by hypotension or by CVP measurements requires more aggressive therapy, but wherever possible plasma electrolyte concentrations should be closely monitored. Because of elevated requirements for glucose in childhood severe malaria (3--6 mg/kg/min) \[[@pmed-0010018-b35]\], there is a need to provide a maintenance fluid replacement rate of about 3 ml/kg/h \[[@pmed-0010018-b07],[@pmed-0010018-b36]\]. In addition, fluid regimens should aim to replace mild fluid deficit within the first 12 to 24 h of admission. The tracer dilution techniques that we have used are expensive and time-consuming and consequently not amenable to large-scale deployment. We took this opportunity to calibrate a much simpler methodology (BIA) to derive TBW and ECW and validated BIA estimations in this population. We are now using BIA measurements to assess much larger numbers of patients (J. Jarvis and S. K., unpublished data). BIA is an excellent noninvasive screening tool that should detect subgroups of children with severe malaria who may be severely volume depleted. ::: {.caption} ###### Patient Summary ::: {#sb1a} ### Background {#sb1a1} Although we have known for many years what causes malaria, how it is passed from person to person by mosquito, and how to treat the infection, more than a million people still die of malaria every year, mostly children under five years living in Africa. Children are affected most because they have not had the chance to develop the resistance to malaria that normally builds up over a lifetime when living in places with malaria. As well as being given specific drug treatments against the malaria, children are also often given fluids into their veins, as they appear dehydrated. ### What Did the Researchers Find? {#sb1a2} The researchers studied children who were sick with malaria and measured how dehydrated the children were. To do the measurements, they used an accepted technique that required injections into the vein, and also a newer, simpler method that used electrodes. Neither technique suggested that any of the children were severely dehydrated. ### What Does This Mean for Patients? {#sb1a3} The study suggests that severe dehydration isn\'t a big problem in children with severe malaria. So it may not be necessary to give lots of fluids into the vein. (While treatment of malaria is still being worked out, malaria can, of course, often be prevented by using insecticide-treated bednets.) ### Are There Any Problems with the Study? {#sb1a4} The new technique for measuring dehydration will need to be assessed in larger studies: this study is small, so the results may not be entirely accurate. ### Where Can I Get More Information? {#sb1a5} The World Health Organization is coordinating many of the initiatives to combat malaria (<http://www.who.int/topics/malaria/en/>). Medicines for Malaria Venture is trying to develop new affordable antimalarial drugs (<http://www.mmv.org/pages/page_main.htm>). We thank the medical, nursing, and laboratory staff of the Albert Schweitzer Hospital and of the Centre Hospitalier de Libreville, especially Ms. K. Engel, Dr. A. Josseaume, Dr. R. Tchoua, Prof. D. Ngaka, and Prof. R. Tchoua for advice and aid in conducting this study. We also thank M. Nestor Obiang, Batchelili Batchelili, Emmanuel Mozogho, and Frankie Mbandinga for their assistance and Martin Hurl of YSI for loaning the YSI2300 analyser. We also thank Prof. E. Ngou-Milama for his aid in this study. Tim Planche was funded by the Special Trustees of St. Georges Hospital, and this work forms a part of his MD thesis. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Citation: Planche T, Onanga M, Schwenk A, Dzeing A, Borrmann S, et al. (2004) Assessment of volume depletion in children with malaria. PLoS Med 1(1): e18. ^2^H~2~O : heavy water adj r^2^ : adjusted r^2^ BIA : bioelectrical impedance analysis CVP : central venous pressure ECW : extracellular water volume ICW : intracellular water volume IQR : interquartile range SEE : standard error of the estimate TBW : total body water volume [^1]: Competing Interests: The authors have declared that no competing interests exist. [^2]: Author Contributions: TP, MK, JFF, MO, PGK, and SK designed the study. TP, SB, and AD analysed patient data; AS analysed the BIA data and constructed the models; AW and LB conducted the analysis of heavy water; and LW analysed the bromide data. TP and SK wrote the first draft of the paper; all authors commented extensively on subsequent drafts.	PubMed Central	2024-06-05T03:55:48.587987	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523837/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e18", "authors": [ { "first": "Timothy", "last": "Planche" }, { "first": "Myriam", "last": "Onanga" }, { "first": "Achim", "last": "Schwenk" }, { "first": "Arnaud", "last": "Dzeing" }, { "first": "Steffen", "last": "Borrmann" }, { "first": "Jean-François", "last": "Faucher" }, { "first": "Antony", "last": "Wright" }, { "first": "Les", "last": "Bluck" }, { "first": "Leigh", "last": "Ward" }, { "first": "Maryvonne", "last": "Kombila" }, { "first": "Peter G", "last": "Kremsner" }, { "first": "Sanjeev", "last": "Krishna" } ] }
PMC523838	Introduction {#s1} ============ Numerous large epidemiological studies have unambiguously demonstrated a strong inverse relationship between the plasma concentration of high-density lipoprotein cholesterol (HDL-c) and the incidence of coronary heart disease (CHD) \[[@pmed-0010019-b01],[@pmed-0010019-b02]\]. Recent attempts to develop therapies aimed at increasing HDL-c as innovative CHD-risk-reducing strategies illustrate the potential of HDL-c as a potent anti-atherogenic mediator \[[@pmed-0010019-b03],[@pmed-0010019-b04],[@pmed-0010019-b05],[@pmed-0010019-b06]\]. Combination antiretroviral therapy (ART) for the treatment of HIV-1 infection has been associated with fat redistribution, insulin resistance, and changes in plasma concentrations of lipids and lipoproteins \[[@pmed-0010019-b07],[@pmed-0010019-b08],[@pmed-0010019-b09]\]. Each of these phenomena is associated with increased CHD risk in the general population. It is not surprising, therefore, that in the setting of HIV-1 infection, increasing exposure to potent combination ART has been demonstrated to be associated with an incremental risk of CHD in a recent prospective study \[[@pmed-0010019-b10]\]. Interestingly, however, the changes in lipids and lipoproteins differ between patients using an ART regimen containing either a protease inhibitor (PI) or a non-nucleoside reverse transcriptase inhibitor (NNRTI). Whereas many of the PI-based regimens are often associated with increased levels of triglycerides (TGs), total cholesterol (TC), and low-density lipoprotein cholesterol (LDL-c) \[[@pmed-0010019-b08],[@pmed-0010019-b09],[@pmed-0010019-b11]\], NNRTI-based regimens importantly differ from PI-based regimens by being associated with marked increases of HDL-c and lesser increases of LDL-c and TGs \[[@pmed-0010019-b12],[@pmed-0010019-b13]\]. Notably, the increases in HDL-c demonstrated with NNRTI-containing ART markedly exceed those that may be induced with any of the currently licensed statins or fibrates \[[@pmed-0010019-b14]\]. Although as yet no clinical data have been generated to support this, these differences between ART regimens raise the expectation that NNRTI-based regimens, particularly in view of their effects on HDL-c, may favourably modify the CHD risk compared with many of the PI-containing regimens. With respect to the two currently commonly used NNRTIs, nevirapine (NVP) and efavirenz (EFV), no detailed comparative data have been reported concerning their effect on plasma lipids and HDL-c in particular. We prospectively analysed lipid and lipoprotein changes in a preplanned substudy of the 2NN trial in which ART-naive patients received stavudine (d4T) and lamivudine (3TC) with the randomly assigned addition of NVP, EFV, or both drugs combined. Methods {#s2} ======= Participants and Treatment Allocation {#s2a} ------------------------------------- The 2NN trial was an open-label study, the main results of which have been published elsewhere \[[@pmed-0010019-b15]\]. Patients enrolled were 16 y of age or older, ART-naive, and had a plasma HIV-1 RNA concentration (pVL) of at least 5,000 copies/ml. Main exclusion criteria were pregnancy or breastfeeding, abnormal laboratory results at screening, the use of immuno-modulating therapy, or anticipated nonadherence. All patients used d4T (40 mg twice daily \[bd\] or 30 mg bd when less than 60 kg) and 3TC (150 mg bd). In addition, patients were randomly allocated to NVP at 400 mg once daily (od), NVP at 200 mg bd, EFV at 600 mg od, or NVP and EFV at 400 mg od and 800 mg od, respectively. Patients were included from 65 different study sites in 17 countries in Asia, Australia, North America, South America, South Africa, and Europe. The 2NN study had been approved by the ethics committees of all participating institutions, and all patients had given written informed consent. The current analyses were preplanned. Only those patients were included who used all components of their allocated treatment for at least 95% of the time during the 48 wk of follow-up (self-reported). Change of d4T and/or 3TC was allowed for reasons of toxicity. Employing such 'on treatment' (OT) analysis, allows the best possible assessment of lipid changes that actually result from differences in regimens. Patients in the NVP-od and NVP-bd groups were combined, given that the virologic efficacy of these treatments was comparable and no differences in risk of virologic failure were observed. Follow-Up and Assessments {#s2b} ------------------------- Plasma samples for prospective determination of lipids and lipoproteins were collected at baseline (before start of treatment) and at weeks 2, 4, 8, 12, 24, 36, and 48. Blood was drawn after a mandatory fast of at least 3 h. The samples were analysed in local laboratories according to predefined protocols. These laboratories were selected by the Virtual Central Laboratory (Zeist, The Netherlands), which selected the laboratories, assured the quality of the analyses and data, and standardised all results. Plasma concentrations of HDL-c, TC, and TGs were assessed by standard enzymatic assays. The concentration of LDL-c was calculated using the Friedewald equation, but only when the concentration of TGs was below 4.5 mmol/l \[[@pmed-0010019-b16]\]. Because the calculation of LDL-c depends on the measured TG concentrations and these TG levels might be biased because of the relatively short mandatory fasting period, we also calculated the non-HDL-c levels. These are considered to be much less influenced by TG levels. The pVL was measured at a central laboratory (LabCorp, Research Triangle Park, North Carolina, United States) using Ultra Sensitive Roche Amplicor 1.5 (Roche Diagnostics, Basel, Switzerland) with a lower limit of quantification of 50 copies/ml. Outcome Measurements {#s2c} -------------------- The primary study outcome was the mean percentage change of HDL-c, TC, TC:HDL-c ratio, non-HDL-c, LDL-c, and TGs between start of allocated treatment and week 48. For each patient at each specific study week we calculated this estimate as concentration at week X minus concentration at baseline divided by concentration at baseline, times 100. Study-week-specific estimates were used for the subsequent analyses. Factors assessed for a possible association with the primary outcome were sex, study region (Asia/Australia, South Africa, South America, Europe/North America), body mass index (BMI) (continuous), increase between start of therapy and week 48 in CD4+ cells (\<100, 100--250, or \>250 cells/mm^3^) or decrease in pVL (\<2.5, 2.5--3.5, or \>3.5 log~10~), and virologic failure during follow-up. Virologic failure was defined as (1) never having obtained a pVL of less than 50 copies/ml or (2) a rebound to two consecutive pVLs of ≥50 copies/ml. A single pVL of ≥50 copies/ml at week 48 was also considered a virologic failure. Statistical Analyses {#s2d} -------------------- The analyses included the NVP and EFV treatment groups only. This choice was made since the results of the main 2NN study clearly showed that the simultaneous use of NVP and EFV shouldn\'t be recommended in clinical practice in view of increased toxicity of this combination in the absence of increased virologic efficacy. The mean percentage changes in lipid concentrations were modelled using a mixed model incorporating repeated measurements. This model handles missing data adequately by estimating the outcome of a specific variable based on the available data given the specified covariate structure. The variables (fixed effects) in the model were tested for significance using the Type III F-statistic. The estimates of a specific level of the fixed effect were modelled using the 'least squared means\' approach. Differences in these estimates between different levels of the variable were tested for significance using the t-statistic. Since the analyses might be biased because of the OT approach or the modelling of data, we performed two sensitivity analyses. The first was an analysis using the same modelling strategy but for an intention-to-treat population including all patients who started their randomised treatment. The second was an analysis using only available data for the OT population, without modelling of data points. Independent risk factors were assessed by multivariable regression analyses. The multivariable analysis included the variable 'treatment group\' and all predefined variables. Interaction between treatment group and a specific variable was assumed at a p-value less than 0.15. A two-sided p-value less than 0.05 was considered statistically significant in the final analyses. The SAS statistical package was used for analyses (version 8.02, SAS Institute, Cary, North Carolina, United States). Results {#s3} ======= Disposition of Patients {#s3a} ----------------------- Of the 1,216 patients included in the 2NN study, 607 were allocated to the NVP treatment group and 400 to the EFV treatment group. Of these, 42 (6.9%) patients in the NVP group and 25 (6.3%) patients in the EFV group did not start their treatment or were considered a 'study entry violator' by a (blinded-to-treatment) independent endpoint committee. These patients were excluded from the analyses. From the remaining patients (565 using NVP and 375 using EFV), only those who remained on their assigned treatment during the follow-up were included in the analyses. This resulted in a final sample size of 417 (68.7%) patients in the NVP group and 289 (72.3%) in the EFV group. All of the included patients had at least one measurement of each lipid parameter and could therefore be used in the statistical models. The baseline characteristics of the subset of patients included in the current analyses are summarised in [Table 1](#pmed-0010019-t001){ref-type="table"}. These baseline characteristics were comparable with those of all patients enrolled in the main 2NN study. ::: {#pmed-0010019-t001 .table-wrap} Table 1 ::: {.caption} ###### Baseline Characteristics of Patients Included in the 2NN Lipid Substudy and the 2NN Main Study ::: ![](pmed.0010019.t001) IQR, interquartile range ::: In the NVP group, 148 of the 565 eligible patients (26.2%) were not included in the OT analyses. Of these 96 (65%) were nonadherent (including patients lost to follow-up while on randomised treatment), 30 (20%) changed their NNRTI to EFV, and 22 (15%) changed their regimen by adding a PI. In the EFV group, 76 of the 375 eligible patients (20.3%) were not included in the OT analyses. Of these 54 (71%) were nonadherent, 17 (22%) changed their NNRTI to NVP, 3 (4%) added a PI, and 2 (3%) added a third nucleoside reverse transcriptase inhibitor. Changes in Lipids and Lipoproteins {#s3b} ---------------------------------- All changes within the treatment groups in lipid and lipoprotein concentration, as well as in TC:HDL-c ratio were statistically significant. The increase of HDL-c was 8.9% (95% confidence interval \[CI\], 0.6--17.1) larger in the NVP treatment group (42.5%) than in the EFV treatment group (33.7%). This was statistically significant (p = 0.036) ([Table 2](#pmed-0010019-t002){ref-type="table"}). In contrast, the increase in TC was smaller in the NVP group (26.9%) than in the EFV group (31.1%), but this difference (−4.2%; 95% CI, −8.7 to 0.4) was not statistically significant (p = 0.073). These changes resulted in a decrease of the TC:HDL-c ratio in the NVP group (−4.1%) compared to an increase in the EFV group (+5.9%; p \< 0.001), and a significantly smaller increase of non-HDL-c in the NVP group (difference, −8.9%; 95% CI, −15.4 to −2.5; p = 0.007). ::: {#pmed-0010019-t002 .table-wrap} Table 2 ::: {.caption} ###### Lipid Concentrations at Baseline and Week 48 and Mean Percentage Change ::: ![](pmed.0010019.t002) All percentage changes within a treatment group were statistically significant ^a^ Units: mmol/l, median (interquartile range) ^b^ Mean percentage change (standard error), modeled by repeated measurements. Mean percentage change was calculated at each specific time point for each individual patient as ((concentration\[week X\] -- concentration\[baseline\]) / concentration\[baseline\]) × 100 ::: The increase of TGs was 28.9% (95% CI, −42.3 to −5.0) smaller in the NVP group (20.1%) than in the EFV group (49.0%; p \< 0.001). The difference in LDL-c increase was not statistically significant (35.4% for NVP group; 40.0% for EFV group; p = 0.378). In the first sensitivity analysis (intention-to-treat population), the increases of HDL-c were slightly lower (41.2% for NVP group; 32.4% for EFV group), just as for TC (26.1% for NVP group, 30.4% for EFV group) and non-HDL-c (24.3% for NVP group, 33.1% for EFV group). The TC:HDL-c ratio showed a smaller decrease for patients taking NVP (−2.6%) but a larger increase for patients taking EFV (+7.2%). The increase in TGs was larger for both patients taking NVP (24.3%) and patients taking EFV (49.3%). The LDL-c increase was somewhat smaller for patients taking NVP (33.1%) but larger for patients taking EFV (47.3%). The difference between patients taking NVP and those taking EFV for HDL-c (8.8%; 95% CI, 1.3−16.3) remained statistically significant, just as the difference in the TC:HDL-c ratio (−9.8%; 95% CI, −14.7 to −4.9), non-HDL-c (−8.8%; 95% CI, −14.6 to −3.0), and TGs (−24.9%; 95% CI, −37.2 to --12.6). Additionally, the difference between NVP and EFV treatment groups became statistically significant for TC (−4.2%; 95% CI, −8.5 to 0.0) and LDL-c (−14.2%; 95% CI, −28.4 to 0.0) compared to the original OT analysis. The second sensitivity analysis (using only available data for the OT population) also showed comparable estimates (data not shown). The increase in HDL-c for patients who started their ART when their HDL-c levels were, according to the National Cholesterol Education Program (NCEP) guidelines, low (\<1.03 mmol/l), normal (1.03--1.55 mmol/l), or high (\>1.55 mmol/l) is reported in [Table 3](#pmed-0010019-t003){ref-type="table"}. In both treatment groups, the majority of patients had a low HDL-c at the start of therapy. These patients showed the largest increase in HDL-c over 48 wk. Even patients with a normal baseline HDL-c level showed statistically significant, marked increases of HDL-c. The effect of baseline HDL-c level on percentage increase was comparable in both treatment groups (interaction, p = 0.409). ::: {#pmed-0010019-t003 .table-wrap} Table 3 ::: {.caption} ###### Increase in HDL-c Stratified by Baseline HDL-c (NCEP Categories) ::: ![](pmed.0010019.t003) ^a^ Mean percentage change (standard error), modeled by repeated measurements. Mean percentage change was calculated at each specific time point for each individual patient as ((concentration\[week X\] -- concentration\[baseline\]) / concentration\[baseline\]) × 100 ::: Multivariable Analysis {#s3c} ---------------------- Factors independently associated with changes in the lipid concentrations were analysed by a multivariable regression analysis ([Table 4](#pmed-0010019-t004){ref-type="table"}). ::: {#pmed-0010019-t004 .table-wrap} Table 4 ::: {.caption} ###### Factors Associated with Percentage Change in Lipid Parameters (Multivariable Analyses) ::: ![](pmed.0010019.t004) ^a^ Units: mmol/l ^b^ Percentage increase (standard error) between baseline and week 48, modelled by repeated measurements. Mean percentage change was calculated at each specific time point for each individual patient as ((concentration\[week X\] -- concentration\[baseline\]) / concentration\[baseline\]) × 100 ^c^ For percentage increase ^d^ A, Asia/Australia; B, South Africa; C, South America; D, Europe/North America ::: Men had a significantly smaller increase of HDL-c, compared to women, but a larger increase of TC. This resulted in an increased TC:HDL-c ratio for men and a decreased ratio for women, while the increase of non-HDL-c was significantly larger in men. The changes in lipid concentrations varied markedly by region. Patients from Asia/Australia and South Africa had the largest decrease in the TC:HDL-c ratio because of an increase of HDL-c that outweighed the increase of TC. Patients from South America, compared to those from other regions, had a significantly smaller HDL-c increase with a comparable TC increase, resulting in an increased TC:HDL-c ratio. Although patients from Europe showed the largest change in HDL-c, the increases in TC and TC:HDL-c ratio were intermediate. A striking finding is the much larger increase of TGs in patients from South America compared to those in patients from other regions, which to a lesser extent was also seen for non-HDL-c. For all lipid concentrations, except TGs, there was a clear pattern of larger increases of lipid levels with larger decreases of pVL over 48 wk. This was also seen when the pVL increase over 48 wk was analysed as a continuous variable. For each log~10~ larger decrease in pVL there was a 4.6% increase in TC (p \< 0.001), a 7.8% increase in HDL-c (p \< 0.001), a 10.2% increase in LDL-c (p \< 0.001), and a 3.6% increase in non-HDL-c (p = 0.002), while the TC:HDL-c ratio declined with 1.6% (p = 0.051). In this analysis, there was also a clear association between pVL decline and change in TGs (6.3% decline per log~10~; p = 0.002). In general, a smaller CD4+-cell increase was associated with a smaller increase in lipid concentration, while increases of more than 250 cells/mm^3^ did not show markedly different effects compared to increases of 100--250 cells/mm^3^. When analysed as a continuous variable, there was no statistically significant association between CD4+-cell increase and change in any of the lipid parameters. BMI was independently associated with increases in all lipid parameters, except TC. Although the increases per unit increase in BMI were statistically significant, the magnitude of increases was rather low. All these factors exhibited a similar effect in both the NVP and the EFV treatment group (no significant interactions). There were, however two exceptions. For changes in TGs, the effect of sex and pVL decrease differed between the treatment groups (interaction, p = 0.005 and p = 0.075, respectively). Men had a significant increase of TGs in both the NVP group (14.1%) and the EFV group (43.3%); women using NVP had no significant TG increase (6.5%), while those using EFV had (15.9%). The effect of pVL decrease on TG increase was quite different for patients taking NVP versus those taking EFV. In the NVP group, the increase in TG concentration was 17.2% for a pVL decrease less than 2.5 log~10~, 16.6% for a decrease between 2.5 and 3.5 log~10~, and −6.1% (denoting a decrease) for a pVL decrease more than 3.5 log~10~. In the EFV group, these estimates were 27.4%, 37.2%, and 26.0%, respectively. Adjusting for the variables included in the multivariable model, the difference between patients taking NVP and those taking EFV in HDL-c increase (9.8%; 95% CI, 3.4−16.3) and decrease of the TC:HDL-c ratio (−11%; 95% CI, −15.1 to −6.8) remained statistically significant. Also, the difference in non-HDL-c increase (−9.5%; 95% CI, −14.6 to −4.4) and TG increase (−27.2%; 95% CI, −38.0 to −16.4) remained statistically significant. The difference in TC increase (−4.4%; 95% CI, −8.0 to −0.8) became statistically significant. The difference between NVP and EFV groups for the increase in LDL-c remained statistically nonsignificant (−6.1%; 95% CI, −14.7 to 2.6). The adjusted increase of HDL-c was 42.3% and 32.4% for patients taking NVP and EFV, respectively (p = 0.003). The adjusted change in TC:HDL-c ratio was −4.3% for patients taking NVP and +6.6% for patients taking EFV (p \< 0.001). These values were 26.6% and 31.0% for TC (p = 0.020), 24.4% and 33.9% for non-HDL-c (p \< 0.001), 17.9% and 45.1% for TG (p \< 0.001), and 35.5% and 41.5% for LDL-c (p = 0.168). These estimates were very similar in the two sensitivity analyses (data not shown). The proportional changes of the different plasma lipid concentrations over 48 wk are graphically depicted in [Figure 1](#pmed-0010019-g001){ref-type="fig"}. ::: {#pmed-0010019-g001 .fig} Figure 1 ::: {.caption} ###### Change in Plasma Concentrations of Lipids and Lipoproteins Adjusted for sex, region, pVL decrease, and CD4+-cell increase. ::: ![](pmed.0010019.g001) ::: Discussion {#s4} ========== Initiation of an ART regimen containing NVP or EFV is accompanied by a significant increase of HDL-c, with concomitant increases of TC, non-HDL-c, TGs, and LDL-c. The proportional increase of HDL-c was significantly larger in the NVP treatment group compared to the EFV treatment group, while the proportional increase of TC, non-HDL-c, and TGs was significantly smaller. In the NVP group, the TC:HDL-c ratio decreased, compared to an increase in the EFV group. These observations are different from what is seen with most PI-based ART regimens, in which higher concentrations of TC, LDL-c, and TGs are reported but without the concurrent higher levels of HDL-c \[[@pmed-0010019-b17],[@pmed-0010019-b18]\]. In contrast to a small randomised study (n = 67) that did not show significant differences between NVP and EFV \[[@pmed-0010019-b19]\], the present study demonstrates a more favourable lipid profile for treatment including NVP than for treatment including EFV in ART-naive patients. HDL-c Increase and NNRTI {#s4a} ------------------------ Increases of HDL-c with the use of NVP or EFV have been described in studies for patients switching from a PI-based regimen to a NNRTI-based regimen \[[@pmed-0010019-b20],[@pmed-0010019-b21]\]. Data for ART-naive patients starting therapy with an NNRTI-based regimen are scarce. Van der Valk et al. reported an increase of HDL-c of 0.44 mmol/l for patients initiating treatment with didanosine, d4T, and NVP in the Atlantic trial \[[@pmed-0010019-b12]\]. Tashima et al. reported an increase of HDL-c of 0.21 mmol/l in patients treated with EFV and either zidovudine plus 3TC, or indinavir \[[@pmed-0010019-b13]\], while Negredo et al. showed an increased HDL-c concentration in therapy-naive patients starting a regimen of didanosine, d4T, and EFV (0.34 mmol/l) \[[@pmed-0010019-b22]\]. The present study showed a clear effect of baseline HDL-c on the proportional increase. The largest increases were seen for patients who had an increased CHD risk based on their low HDL-c level (\<1.03 mmol/l) according to NCEP guidelines. But also patients with a normal HDL-c level, who are not at an increased CHD risk, showed marked increases in HDL-c. This baseline effect can likewise be distilled from the other studies, where those with the lowest baseline value (0.93 mmol/l; Atlantic study \[[@pmed-0010019-b12]\]) showed the largest HCL-c increase, while the smallest increase was seen in the study with the highest baseline value (1.23 mmol/l; Tashima study \[[@pmed-0010019-b13]\]). The study by Negredo et al. \[[@pmed-0010019-b21]\], which included patients with similar baseline HDL-c levels as in the present study, showed an increase of HDL-c comparable to that in the present study (0.34 and 0.36 mmol/l, respectively). The much more modest HDL-c-increasing effect of statins likewise shows such a correlation with baseline level in patients without HIV-1. One may postulate that the HDL-c increase merely reflects an adequate suppression of HIV-1 infection ('return towards normal'). In support, a larger decrease in pVL was associated with a larger increase of HDL-c in the present study. However, the magnitude of the HDL-c increase was only slightly different for patients experiencing virologic failure during these 48 wk (29.5%) and those with complete suppression (34.0%; p = 0.161), and the increases of HDL-c remained statistically significant even after adjustment for pVL decrease. Riddler et al. compared changes in lipid concentrations before seroconversion for HIV-1, initiation of ART, and during ART in patients using different ART regimens, which all but one included a PI \[[@pmed-0010019-b23]\]. The period between seroconversion and start of ART was characterised by decreases in TC, LDL-c, and HDL-c. Between initiation of ART and the first follow-up visit (mean, 1.3 years), the concentrations of TC and LDL-c increased again to levels that did not differ significantly from before seroconversion. Such a 'return to normal' as a result of ART was not seen for HDL-c. The reported increases by Riddler et al. in TC (0.88 mmol/l) and LDL-c (0.41 mmol/l) were of a comparable magnitude to that found for patients taking NVP in this present study (0.97 and 0.55 mmol/l, respectively). However, the HDL-c increase was more than ten times smaller (0.03 mmol/l) in the Riddler et al. study than the 0.36 mmol/l observed in patients taking NVP in the present study, while the mean HDL-c values at which ART was started were comparable in the two studies (1.04 and 1.0 mmol/l, respectively). This indicates that although at least part of the change in TC and LDL-c may reflect a 'return towards normal', the magnitude of the HDL-c increase observed in our study must have occurred through additional mechanisms. Since we have no information on the antiretroviral efficacy of the regimens used in the Riddler et al. study, we have to consider that the reported differences between the Riddler et al. study and the present study might be partly due to differences in HIV-1 suppression. However, the type of PI-based regimens used in the Riddler et al. study and the long-term adequate adherence by the patients make large differences in antiretroviral efficacy unlikely. We are currently conducting studies to unravel whether NVP possibly stimulates synthesis of the most important apolipoprotein of HDL-c, apoAI, or alternatively, for instance, decreases the clearance of HDL-c particles. Several studies have convincingly shown that an HDL-c increase is associated with a significant decrease in CHD mortality independent of changes in LDL-c \[[@pmed-0010019-b01],[@pmed-0010019-b02]\]. Overall, extrapolation of these studies indicates that a 0.025-mmol/l increase in HDL-c is expected to be associated with a 2%--3% reduction in CHD risk, while an increase of 1.0 mmol/l in LDL-c will increase the CHD risk by 25%. The mean absolute increases in HDL-c and LDL-c were 0.36 and 0.54 mmol/l, respectively, for patients taking NVP, and 0.24 and 0.65, respectively, for patients taking EFV. It can therefore be estimated that, taking the observed effects on both HDL-c and LDL-c into account, the reduction in CHD risk would be 15% for patients taking NVP and 3% for patients taking EFV compared to ART regimens that do not include NNRTIs. Although the differences in absolute concentrations of HDL-c and LDL-c may seem modest when comparing the NVP and EFV treatment groups, the combined effect of these changes on CHD risk seems marked. It should be emphasised that these are theoretical estimates, which do not take into account that increases in TGs would be expected to have an opposite effect on CHD risk. The increase in this last parameter is, however, smaller for patients taking NVP than for patients taking EFV. Furthermore, we do not have information on the presence of conventional risk factors for CHD. The actual effect of the lipid changes associated with particular ART regimens on CHD can only be substantiated by clinical endpoint studies. Changes in TGs, TC, and LDL-c {#s4b} ----------------------------- The data indicate that EFV might have a more detrimental effect on TG levels than NVP. That EFV indeed can be associated with an increase in TGs was shown in two studies, in which sporadical hypertriglyceridaemia was reported in patients starting an ART regimen with EFV but without d4T \[[@pmed-0010019-b24],[@pmed-0010019-b25]\]. A difference between NVP and EFV treatment with respect to the TG effect is also in line with a study by Negredo et al. \[[@pmed-0010019-b20]\]. In this study patients were randomised to either continue their successful PI-based regimen or to change to an NVP-based or an EFV-based regimen. Only patients switching to the NVP regimen showed a significant decrease in TG levels. The proportional increase in TG levels with both NVP and EFV treatment seems large, but the median absolute TG level at week 48 was still low in both treatment groups (1.2 mmol/l and 1.4 mmol/l, respectively). In the NCEP guidelines, a TG concentration below 1.69 mmol/l is still considered normal \[[@pmed-0010019-b26]\]. The increase in TG level is therefore probably not clinically meaningful. The differences between patients taking NVP and those taking EFV in changes in TC as well as TGs are unlikely to be explained by the concurrent use of d4T. In both treatment groups, the percentage of patients who used d4T as part of their regimen throughout follow-up was high (96% for the NVP treatment group and 98% for the EFV treatment group). This high rate of d4T use might, however, be responsible for the impression that the increases in TC and TGs seem to continue or even accelerate towards the end of the study period, as opposed to a somewhat declining effect of treatment on HDL-c after 24 wk. A possible explanation for this may be the gradual, progressive worsening of fat redistribution or lipodystrophy that one would expect to occur in this continuously d4T-exposed patient population. Both the incidence and severity of lipodystrophy are particularly increased with d4T-containing ART regimens, and lipodystrophy has been reported to be associated with increased TC and TG levels \[[@pmed-0010019-b27],[@pmed-0010019-b28],[@pmed-0010019-b29],[@pmed-0010019-b30],[@pmed-0010019-b31],[@pmed-0010019-b32]\]. It is therefore conceivable that the lipid changes in the second part of the study represent a combined effect of the NNRTI used and a superimposed effect resulting from gradually worsening fat redistribution. Due to the relatively short mandatory fasting period (3 h), the measured TG concentration might be biased, possibly even more so given that HIV-1 infection may be associated with reduced TG clearance following food intake \[[@pmed-0010019-b33]\]. As a consequence, the estimates of calculated LDL-c might be biased. TG levels influence changes in non-HDL-c less. The fact that in the present study the increases of non-HDL-c, LDL-c, and TGs are all smaller for patients taking NVP than for patients taking EFV suggests that the LDL-c and TG results are valid despite the potentially short mandatory fasting periods. ART and CHD {#s4c} ----------- The relationship between ART and CHD has been the subject of several studies, based on either clinical or validated surrogate endpoints (like arterial intima-media thickness \[[@pmed-0010019-b34]\] or endothelial wall function \[[@pmed-0010019-b35]\]). Studies examining intima-media thickness in patients with HIV-1 treated with ART, or more specifically with PI-based ART, remain inconclusive as to whether ART use induces accelerated intima-media thickening \[[@pmed-0010019-b36],[@pmed-0010019-b37],[@pmed-0010019-b38],[@pmed-0010019-b39]\]. PI use may have a detrimental effect on endothelial function in vivo, or accelerate foam cell formation in vitro \[[@pmed-0010019-b40],[@pmed-0010019-b41]\]. In a retrospective clinical endpoint study including almost 37,000 patients, Bozzette et al. reported no relation between ART use and hospital admission for cardiovascular events \[[@pmed-0010019-b42]\], a finding confirmed in the 'Kaiser Permanente' cohort \[[@pmed-0010019-b43]\]. However, the large prospective 'data collection on adverse events of anti-HIV drugs\' (D:A:D) study, specifically designed to identify the extent to which ART may be associated with increased CHD risk, did show that every additional year of ART use was associated with a 26% increased risk of myocardial infarction \[[@pmed-0010019-b10]\]. The latter resembles the results from other retrospective and prospective studies \[[@pmed-0010019-b44],[@pmed-0010019-b45],[@pmed-0010019-b46]\]. The relatively high prevalence of known CHD risk factors in patients with HIV-1, especially smoking \[[@pmed-0010019-b47],[@pmed-0010019-b48]\], complicates interpretation of the relation between ART use and CHD. None of these studies allows definitive conclusions to be made about the potentially different degree of risk associated with particular ART regimens. Limitations and Possible Biases {#s4d} ------------------------------- The selection of patients remaining on their allocated treatment for the full 48 wk might have introduced sampling bias with inflated treatment effects. The reported estimates from this OT analysis were very similar to the estimates from the intention-to-treat analysis. This is caused by the fact that only a few patients who were not included in the OT analyses changed their drug regimen by adding a PI that could potentially influence the lipid estimates. The majority remained on their randomised treatment but were insufficiently adherent (or lost to follow-up on their original regimen) to meet the criteria for being eligible for the OT analysis. Patients replacing their assigned NNRTI by the NNRTI from the other treatment group of the study would have little effect on the lipid estimates, since both NNRTIs show changes in lipid concentrations, which go in the same direction. Another possible reason for the similarity between these two analyses could be the relatively late timing of treatment changes (mean of 75 d for patients taking NVP and 95 d for patients taking EFV). The fact that the second sensitivity analysis also showed comparable results indicates that modelling of data did not affect the lipid estimates. A limitation of the present study is the lack of data on conventional CHD risk factors like smoking. The 2NN being a randomised study, it may be expected that these and other confounding variables are equally distributed over the treatment groups. Possible residual confounding, however, cannot be excluded, and the results should therefore be interpreted cautiously. Conclusion {#s4e} ---------- While awaiting the results of future studies, the less atherogenic lipid profile of patients taking NVP in comparison to those of patients taking EFV may be among the various factors to consider when selecting the most appropriate initial ART regimen, particularly for those patients with HIV-1 with a significant a priori CHD risk. Including such a consideration seems warranted, since treatment of the ART-induced lipid changes with currently licensed lipid-lowering agents is not without problems. Most of the available statins except pravastatin and fluvastatin, are metabolised through cytochrome isoenzyme CYP3A4, just as the PIs and NNRTIs are, providing concern for potential drug--drug interactions. Furthermore, statin therapy in patients using a PI-based regimen is in general not able to reduce the lipid concentrations to normal levels \[[@pmed-0010019-b49],[@pmed-0010019-b50]\]. Studies on the effectiveness of gemfibrozil in patients on a PI-based regimen show conflicting results \[[@pmed-0010019-b51],[@pmed-0010019-b52],[@pmed-0010019-b53]\]. Finally, the introduction of yet another type of medication in a patient population that often needs to use not only ART but also a considerable amount of concomitant medication might jeopardise treatment adherence, which is of crucial importance for the sustained success of ART treatment. Use of the novel PI atazanavir may also be considered an attractive option in patients at high risk of CHD, given that it is associated with markedly smaller increases of TC, LDL-c, and TGs compared to previously available PI-based regimens \[[@pmed-0010019-b54],[@pmed-0010019-b55]\], but it lacks the concurrently large increase in HDL-c seen with NNRTI-based regimens. The reported increase of HDL-c concentration with NNRTI use is far greater than that seen with conventional lipid-lowering drugs and is of a similar magnitude as the HDL-c increases reported with the most powerful HDL-c-increasing drugs that are currently in clinical development \[[@pmed-0010019-b06],[@pmed-0010019-b56]\]. Asztalos et al. reported the effect on HDL-c for five major statins in patients with CHD \[[@pmed-0010019-b57]\]. They concluded that the HDL-c increase was between 4% (simvastatin) and 11% (pravastatin and lovastatin). Treatment with fluvastatin or lovastatin proved to be most effective in patients with a low baseline HDL-c level \[[@pmed-0010019-b56],[@pmed-0010019-b58]\]. Clinical trials assessing the effects of fibrates reported an HDL-c increase of 6% and 11% for gemfibrozil \[[@pmed-0010019-b02],[@pmed-0010019-b59]\], and 18% for bezafibrate \[[@pmed-0010019-b60]\]. Unravelling the mechanism or mechanisms by which NVP and EFV raise HDL-c could contribute to the development of novel interventions aimed at increasing HDL-c and thereby ultimately to reducing CHD risk in the population at large. ::: {.caption} ###### Patient Summary ::: ### {#sb1a} #### Why Did the Researchers Do the Study? {#sb1a1} Drugs used to treat HIV (antiretroviral drugs) help patients to live longer, but they can also have some serious side effects. For example, the longer people take them, the higher the risk they\'ll get heart disease. Why? Part of the reason is that many---though not all---antiretroviral drugs cause changes in cholesterol levels in the bloodstream (an increase in the amount of "bad" cholesterol and a reduction in the amount of "good" cholesterol). Two of the most commonly prescribed antiretroviral drugs are nevirapine and efavirenz. Previous smaller studies showed that treatment with either of the drugs could increase the amount of "good" cholesterol. The researchers now wanted to directly compare these drugs to find out what effect they had on patients\' cholesterol levels in a much larger group of patients. #### What Did the Researchers Do? {#sb1a2} The scientists studied adults with HIV who had never previously taken antiretroviral drugs. All of the patients then took "triple therapy"---a combination of three antiretroviral drugs. Some of the patients took nevirapine as part of their triple therapy, whereas some took efavirenz. The researchers took blood samples regularly for almost a year and measured patients\' cholesterol levels. #### What Did the Researchers Find? {#sb1a3} They confirmed that both nevirapine and efavirenz indeed have a beneficial effect on patients\' cholesterol levels. They both increase the amount of "good" cholesterol in the bloodstream. The increase was higher with nevirapine than with efavirenz. #### What Does This Study Mean for Patients? {#sb1a4} If your treatment includes nevirapine or efavirenz (particularly nevirapine), this can raise your level of "good" cholesterol. The results of the study may be especially important if you are already at risk for heart disease. In other words, if you have risk factors for heart disease---high blood pressure, diabetes, heart disease running in your family, being a smoker---it may be beneficial for your HIV medications to include nevirapine. If you smoke, you can lower your risk of heart disease by quitting. If you have high blood pressure or diabetes, treating these conditions can also lower your risk of heart disease. #### What Are the Problems with the Study? {#sb1a5} Although the researchers showed that nevirapine and efavirenz can have a beneficial effect on cholesterol levels, they haven\'t actually shown that this reduces patients\' risk of getting heart disease. The study was funded by the company that produces nevirapine. In theory this could have affected the results (research has shown that company-sponsored studies are more likely to produce results favorable to the company than studies without sponsorship). The study, however, was carried out by a network of independent investigators who state that the company had no influence on the reporting of the results. #### Where Can I Get More Information? {#sb1a6} You can get more information on HIV and its treatment from the Terrence Higgins Trust ([www.tht.org.uk](www.tht.org.uk)), AIDS.ORG ([www.aids.org](www.aids.org)), and The Body ([www.thebody.com](www.thebody.com)). The study was investigator initiated and financially supported by Boehringer Ingelheim. This company had a nonbinding input on issues of study design and analyses, which did not lead to any significant influence on the resulting design and analyses. The company was allowed to provide comments on the manuscript in progress, but had no influence on reporting of the data or the decision to publish. All investigators, as well as Boehringer Ingelheim, had full access to the data after official closure of the database. Investigators ============= Argentina: H. Laplumé, M. B. Lasala, M. H. Losso, E. Bogdanowicz, R. Lattes, A. Krolewiecki, C. Zala, C. Orcese, S. Terlizzi, A. Duran, J. Ebensrteijn. Australia: M. Bloch, O. Russell, D. B. Russell, N. R. Roth, B. Eu, D. Austin, A. Gowers, D. Quan. Belgium: J. Demonty, R. Peleman, B. Vandercam, D. Vogelaers, B. van der Gucht, F. van Wanzeele, M. M. Moutschen. Brazil: R. Badaro, B. Grinsztejn, M. Schechter, D. Uip, E. N. Netto, S. S.Coelho, F. Badaró, J. H. Pilotto, A. Schubach, M. L. Barros, O. H. M. Leite, C. R. V. Kiffer, C. T. Wunsch, D. Nunes, A. Catalani, R. de Cassia Alves Lira, T. J. Dossin, M. T. D\'Alló de Oliveira, S. Martini. Canada: B. Conway, J. J. de Wet, J. S. G. Montaner, C. Murphy, B. Woodfall, P. Sestak, P. Phillips, V. Montessori, M. Harris, A. Tesiorowski, B. Willoughby, R. Voigt, J. Farley, R. Reynolds, S. Devlaming. France: J. M. Livrozet, W. Rozenbaum, D. Sereni, M. A.Valantin, C. Lascoux, B. Milpied, C. Brunet, E. Billaud, A. Huart, V. Reliquet, M. F. Charonnat, M. Sicot, J. L. Esnault, L. Slama. Germany: S. Staszewski, M. Bickel. Greece: M. K. Lazanas, N. Stavrianeas, N. Mangafas, I. Zagoreos, S. Kourkounti, V. Paparizos, C. Botsi. Ireland: S. Clarke, E. Brannigan, N. Boyle. Italy: A. Chiriani, F. Leoncini, F. Montella, L. Francesco, S. Ambu, A. Farese, M. Gargiulo, F. Di Sora, F. Lavria, F. Folgori. Poland: M. Beniowski, A. Boron-Kaczmarska, W. Halota, D. Prokopowicz, D. B. Bander, M. L. P. Leszuzyszyn-Pynka, A. W. Wnuk, E. Bakowska, P. Pulik, R. Flisiak, A. Wiercinska-Drapalo, E. Mularska, A. Witor. Portugal: F. Antunes, R. S. E. Sarmento, M. Doroana, A. A. Horta, O. Vasconcelos. South Africa: S. M. Andrews, C. B. Huisamen, D. Johnson, O. Martin, L.-G. Bekker, G. Maartens, D. Wilson, C. J. Visagie, N. J. David, M. Rattley, E. Nettleship, D. J. Martin, V. Keyser, T. M. Moraites, M. A. Moorhouse, J. A. Pitt, C. J. Orrell, C. Bester, R. Parboosing, P. Moodley, V. Gathiram, D. Woolf. Switzerland: E. Bernasconi, L. Magenta. Thailand: P. Cardiello, E. Kroon, C. Ungsedhapand. United Kingdom: M. Fisher, E. G. L. Wilkins, E. Stockwell, J. Day, R. S. Daintith, N. Perry, C. Timaeus, J. McIntosh-Roffet, A. Powell, M. Youle, M. Tyrer, S. Madge, A. Drinkwater, Z. Cuthbertson, A. Carroll. United States: S. Becker, H. Katner, D. Rimland, M. S. Saag, M. Thompson, M. Witt, M. M. Aguilar, A. LaVoy, M. Illeman, M. Guerrero. Data Safety and Monitoring Board: J. Gatell (chair), E. Belsey, B. Hirschel. Project management: L. Dam, A. Potarca, M. Cronenberg, L. Kreekel. Data management: R. Meester, J. Khodabaks, H.-J. Botma, N. Esrhir, I. Farida, M. Feenstra, K. Jansen, A. Klotz, M. Mulder, G. Ruiter. Laboratory: C. B. Bass, E. Pluymers, E. de Vlegelaer (Labcorp), R. Leeneman (Virtual Central Laboratory). Boehringer Ingelheim: H. Carlier, E. van Steenberge, P. Robinson. Citation: van Leth F, Phanuphak P, Stroes E, Gazzard B, Cahn P, et al. (2004) Nevirapine and efavirenz elicit different changes in lipid profiles in antiretroviral-therapy-naive patients infected with HIV-1. PLoS Med 1(1): e19. 3TC : lamivudine ART : antiretroviral therapy bd : twice daily BMI : body mass index CHD : coronary heart disease CI : confidence interval d4T : stavudine EFV : efavirenz HDL-c : high-density lipoprotein cholesterol LDL-c : low-density lipoprotein cholesterol NCEP : National Cholesterol Education Program NNRTI : non-nucleoside reverse transcriptase inhibitor NVP : nevirapine od : once daily OT : on treatment PI : protease inhibitor pVL : plasma HIV-1 RNA concentration TC : total cholesterol TG : triglyceride [^1]: Competing Interests: FvL has received travel grants and honoraria for presentations from Boehringer Ingelheim and travel grants from GlaxoSmithKline. PP has received research grants and honoraria for speaking from GlaxoSmithKline, Bristol-Myers Squibb, Roche, Merck Sharp and Dohme, and Boerhinger Ingleheim. BG has received grants and honoraria from Abbott Pharmaceuticals, GlaxoSmithKline, Bristol-Myers Squibb, Gilead, and Boehringer Ingleheim. FR has received research grants and honoraria from GlaxoSmithKline, Roche, Bristol-Myers Squibb, Abbott, and Boehringer Ingleheim. RW is conducting clinical research supported by GlaxoSmithKline, Boehringer Ingelheim, Pfizer, and Bristol-Myers Squibb. MB is currently conducting clinical studies supported by Boehringer Ingelheim, Bristol-Myers Squibb, and Merck. CK has received grants and honoraria for speaking from Bristol-Myers Squibb, GlaxoSmithKline, Boehringer Ingelheim, Bayer, and Roche. MS has received research grants, travel grants, and honoraria for speaking from Abbott, Bristol-Myers Squibb, Boehringer Ingelheim, GlaxoSmithKline, Gilead, Merck, and Roche. RLM has received honoraria for consultancies from Boehringer Ingelheim and Bristol-Myers Squibb. AH has received honoraria for presentations from Boehringer Ingleheim. DBH is employed by Boehringer Ingelheim, the manufacturer of one of the trial medications. JMAL has received honoraria as an advisor for GlaxoSmithKline, Boehringer Ingelheim, Bristol-Myers Squibb, Roche, Merck Sharp and Dohme, Schering-Plough, Bayer, Shire Pharmaceuticals, Agouron/Pfizer, and Virco/Tibotec. PR, in the last five years, has received honoraria for speaking engagements from Boehringer Ingelheim. JMAL is a member of the editorial board of PLoS Medicine. [^2]: Author Contributions: FvL, ES, DBH, JMAL, and PR designed the study. FvL and DBH analysed the data. PP, BG, PC, FR, RW, MB, CK, MS, RLM, and AH enrolled patients. FvL wrote the first draft of the paper. All authors contributed to the writing of the final version of the manuscript.	PubMed Central	2024-06-05T03:55:48.590665	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523838/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e19", "authors": [ { "first": "Frank", "last": "van Leth" }, { "first": "Prahpan", "last": "Phanuphak" }, { "first": "Erik", "last": "Stroes" }, { "first": "Brian", "last": "Gazzard" }, { "first": "Pedro", "last": "Cahn" }, { "first": "François", "last": "Raffi" }, { "first": "Robin", "last": "Wood" }, { "first": "Mark", "last": "Bloch" }, { "first": "Christine", "last": "Katlama" }, { "first": "John J. P", "last": "Kastelein" }, { "first": "Mauro", "last": "Schechter" }, { "first": "Robert L", "last": "Murphy" }, { "first": "Andrzej", "last": "Horban" }, { "first": "David B", "last": "Hall" }, { "first": "Joep M. A", "last": "Lange" }, { "first": "Peter", "last": "Reiss" } ] }
PMC523839	Chronic obstructive pulmonary disease (COPD) is a global epidemic of major proportions that is predicted to become the third most common cause of death and fifth most frequent cause of chronic disability by 2020. In developed countries it is mainly caused by cigarette smoking, but the reasons why only a proportion (10%--20%) of smokers develop progressive airflow limitation is currently unknown. The disease is characterized by a chronic inflammatory process predominantly in the small airways and lung parenchyma, with increased numbers of macrophages, neutrophils, and T lymphocytes \[[@pmed-0010020-b1]\]. The difference between smokers without COPD and smokers with COPD appears to be the intensity rather than the nature of the inflammatory process. This inflammation in the small airways is associated with fibrosis and increases with the severity of airflow limitation \[[@pmed-0010020-b2]\], which has led to the view that COPD represents an amplification of the normal inflammatory response to inhaled irritants such as cigarette smoke. T Lymphocytes in COPD {#s2} ===================== T lymphocytes were first reported to be increased in patients with COPD by Finkelstein and colleagues, who showed a correlation between the number of T lymphocytes/mm^3^ of lung and the extent of emphysema \[[@pmed-0010020-b3]\]. It was later shown that both CD4^+^ (T helper) and CD8^+^ (suppressor/cytotoxic) T cells were increased in the airways and lung parenchyma of patients with COPD, with a predominance of CD8^+^ cells \[[@pmed-0010020-b4],[@pmed-0010020-b5]\]. This is in contrast to the findings in asthma, in which there is a predominance of CD4^+^ cells, which are predominantly of the T helper 2 (Th2) pattern, with increased expression of interleukin (IL)-4, IL-5, and IL-13 (see Glossary), and which are associated with an increased number of eosinophils. In smokers who develop COPD there appears to be activation of adaptive immunity, with the infiltration of CD8^+^ and CD4^+^ cells in the alveolar walls and small airways and---in patients with the most severe disease---the presence of lymphoid follicles that contain a core of B lymphocytes surrounded by T cells \[[@pmed-0010020-b2]\]. This activation presumably follows on from the initial and then sustained innate immune response characterized by increased numbers of macrophages and neutrophils; it may involve the migration of dendritic cells from the epithelium to the local lymph nodes and presentation of antigenic substances to T cells, resulting in clonal expansion of CD4^+^ and, to an even greater extent, CD8^+^ cells. The study by Grumelli et al. (2004) published in this issue of PLoS Medicine takes the story forward \[[@pmed-0010020-b6]\]. The CD4^+^ and CD8^+^ cells appear to be fully activated, as they would be after being presented with antigens, and they show predominantly a T helper 1 (Th1)/cytotoxic T 1 (Tc1) pattern, with increased expression of interferon-γ (IFN-γ) and Th1 chemokines. This is consistent with the recent demonstration of increased expression of IL-12 in bronchial biopsies of patients with COPD and activation of the transcription factor STAT-4 in T cells, subsequent STAT-4 nuclear translocation, and IFN-γ gene induction, and thus a Th1 commitment in the T cells \[[@pmed-0010020-b7]\]. As well as producing the cytokines IL-2 and IFN-γ, Th1 and Tc1 cells also express the chemokine receptor CXCR3 and the ligands that activate this receptor, IFN-γ inducible protein 10 (IP-10, CXCL10), monokine induced by IFN-γ (CXCL9), and IFN-inducible T cell α chemoattractant (CXCL11). There is an increase in the expression of IP-10 in the airways of patients with COPD and an increase in the number of CXCR3^+^ cells \[[@pmed-0010020-b8]\]. CXCR3 is expressed on Th1/Tc1 cells, macrophages, and epithelial cells. Release of CXCR3-activating chemokines would attract Th1 and Tc1 cells into the lungs, and these cells then release IFN-γ, which releases more CXCR3 chemoattractants. This results in a self-perpetuating loop that may lead to accumulation of activated Th1 and Tc1 cells in the peripheral lung ([Figure 1](#pmed-0010020-g001){ref-type="fig"}). ::: {#pmed-0010020-g001 .fig} Figure 1 ::: {.caption} ###### In Emphysema, a Self-Perpetuating Loop May Lead to Accumulation of Activated Th1/Tc1T Cells in the Peripheral Lung ::: ![](pmed.0010020.g001) ::: Role of Cytotoxic T Cells {#s3} ========================= It is likely that Th1 cells are the major source of IFN-γ in the lungs of patients with COPD and therefore drive and maintain the T cell response and promote an "immune inflammation" with neutrophils and macrophages. However, it is the role of Tc1 cells that is of particular interest, as these cells are cytotoxic to epithelial cells through the release of granzymes and perforins, which induce apoptosis. Increased concentrations of perforins have recently been reported in the sputum of patients with COPD \[[@pmed-0010020-b9]\]. In support of this idea there is an increase in the apoptosis of alveolar cells in the lungs of patients with COPD, and this is correlated with the number of CD8^+^ cells and the severity of emphysema \[[@pmed-0010020-b10]\]. T Cell Perpetuation {#s4} =================== The T cell inflammatory response appears in mild COPD but increases markedly with disease severity. It is possible that the initial immune response becomes self-perpetuating because of endogenous autoantigens resulting from inflammatory and oxidative lung injury. There are also antigens in tobacco, but the inflammatory response appears to become independent of smoking status, and there is intense inflammation even in patients who stopped smoking many years previously \[[@pmed-0010020-b2]\], as seen in the present study by Grumelli et al. \[[@pmed-0010020-b6]\]. Another possibility is that this chronic immune response is driven, or at least maintained, by chronic infection of the respiratory tract often seen in patients with severe disease, in which there is increased colonization of the lower airways. These infections could act as co-stimulators, or by antigenic mimicry or as polyclonal activators they could provide a persisting antigenic stimulus and maintain the inflammatory process. Further studies on T cell receptor usage and expression of surface markers may give further clues as to the driving mechanisms for the increased Th1 and Tc1 cells in COPD. Proteases {#s5} ========= COPD is characterized by destruction of the lung parenchyma and loss of elastin due to elastolytic enzymes, such as neutrophil elastase and certain matrix metalloproteinases (MMPs). The predominant MMP in COPD appears to be MMP9, which is released in much larger amounts from alveolar macrophages of patients with COPD than from those of smokers without the disease \[[@pmed-0010020-b11]\]. The study by Grumelli et al. showed that CXCR3 ligands led to the expression of the elastolytic enzyme MMP12 in alveolar macrophages and that this process was increased in the lungs of patients with COPD. This finding provides a neat link between T cells and alveolar destruction, but is discrepant with other data that have failed to show significant MMP12 release from macrophages of patients with COPD \[[@pmed-0010020-b11]\]. Therapeutic Implications {#s6} ======================== There are currently no treatments that reduce the relentless progression of COPD, and none that have significant anti-inflammatory effects. However the recognition that an adaptive immune T cell response, most likely driven by antigens, may play an important pathophysiological role in the pathogenesis of COPD has important therapeutic implications. It is possible that T cell inhibitory strategies, such as the use of immunosuppressants, might be effective, although side effects may be a problem, and there is particular concern about increasing the risk of bacterial infection. Another approach might be to block the trafficking of Th1 and Tc1 cells to the lungs by blocking CXCR3, and there is now a search for small-molecule inhibitors of these receptors. Inhibition of IFN-γ signaling might be another approach. The mounting evidence implicating T cells, and thus an adaptive immune response, as an important component of the inflammation in COPD is overwhelming. A better understanding of the immune mechanisms involved in COPD is important, since it might lead us to new and more effective therapeutic approaches to this important disease. ::: {.caption} ###### Glossary ::: CD4^+^ (helper) T cell: T lymphocyte that enhances the inflammatory response CD8^+^ (cytotoxic/suppressor) T cell: T lymphocyte that suppresses the inflammatory response CXCR3: Chemokine receptor that is selectively activated by IP-10, monokine induced by IFN-γ, and IFN-inducible T cell chemoattractant Cytotoxic (Tc1) cell: T cell that is characterized by secretion of INF-γ Granzyme: Enzyme released by cytotoxic T cells Interferon-γ inducible protein 10 (IP-10, CXCL10): Chemokine of 10 kDa that selectively activates CXCR3 Interferon-inducible T cell γ chemoattractant (I-TAC, CXCL11): Chemokine that selectively activates CXCR3 Interferon-γ (IFN-γ): Protein secreted by Th1 and Tc1 cells Interleukin-4 (IL-4): Protein secreted by Th2 cells that is important in increasing IgE secretion Interleukin-5 (IL-5): Protein secreted by Th2 cells that is important for eosinophilia Interleukin-12 (IL-12): Protein secreted by antigen-presenting cells that promotes differentiation of Th1 cells Interleukin-13 (IL-13): Protein secreted by Th2 cells that is important for IgE secretion Matrix metalloproteinase (MMP): Proteolytic enzyme that degrades connective tissue MMP9, MMP12: MMPs that destroy elastin fibers Monokine induced by interferon-γ (MIG, CXCL9): Chemokine that selectively activates CXCR3 Neutrophil elastase: Enzyme released from neutrophils that destroys elastin fibers Perforin: Protein released by cytotoxic T cells that induces apoptosis STAT-4: Transcription factor specifically activated by IL-1 T helper 1 (Th1) cell: T lymphocyte that is characterized by secretion of INF-γ T helper (Th2) cell: T lymphocyte that is characterized by increased secretion of the cytokines IL-4, IL-5, and IL-13; characteristically increased in allergic inflammation Citation: Barnes PJ, Cosio MG (2004) Characterization of T lymphocytes in chronic obstructive pulmonary disease. PLoS Med 1(1): e20. COPD : chronic obstructive pulmonary disease IFN-γ : interferon-γ IL : interleukin IP-10 : interferon-γ inducible protein 10 MMP : matrix metal-loproteinase Tc1 : cytotoxic T 1 Th1 : T helper 1 Th2 : T helper 2 [^1]: Peter J. Barnes is professor and head of thoracic medicine at the National Heart and Lung Institute, Imperial College, London, United Kingdom. Manuel Cosio is professor in the Department of Medicine and research director at the Meakins Christie Laboratories, McGill University, Montreal, Canada. [^2]: Competing Interests: The authors declare that they have no competing interests.	PubMed Central	2024-06-05T03:55:48.594889	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523839/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e20", "authors": [ { "first": "Peter J", "last": "Barnes" }, { "first": "Manuel G", "last": "Cosio" } ] }
PMC523840	Today the possibilities for a medical journal are almost limitless. The first medical journals reflected the needs of a closed group of doctors. But medicine, its place in the world, and the dissemination of information have changed utterly. So in starting afresh, what should a new medical journal retain, and what should it ditch? Most obviously, we should throw out the old way of disseminating information. In today\'s electronic age, it is no more difficult, and it is only minimally more costly, to provide access to one million people than it is to one person. So the revolutionary idea of anyone being able to read any article is possible. This idea---open access---which completely challenges the old subscription-based publishing model, is the driving force behind the launch of PLoS Medicine. You can download and distribute articles without restrictions (feel free to make a thousand copies, translate articles into other languages, put articles into books---just give the author proper credit). We have also changed the way we involve the academic community in our journal. Our large global editorial board reflects the diversity of medicine today and is intimately involved in what we do. In particular, members of the editorial board are a crucial part of our peer review process. As academic editors they, along with a senior editor at the journal, take research papers through the peer review process in a way that we believe provides the most constructive and fair review. We are delighted that members of our editorial board have also shown their support for our journal by submitting papers to us, even before we launched. What will we publish? The research article on malaria in this issue reflects our priority of publishing papers on diseases that take the greatest toll on health globally. But we will also publish papers reporting a substantial advance in any specialty, whether that advance is in public health, such as the paper on the global burden of disease; drug effects, such as the paper on the effect of HIV drugs on lipids; or the molecular understanding of disease, such as the paper dissecting out the immune responses in lung disease caused by smoking. A good general medical journal should also be a place where the global medical community can discuss together what matters to them. The magazine section of PLoS Medicine will be devoted to comment, lively debate, and diverse opinions, in particular giving neglected voices and diseases a place in the limelight. In this issue\'s magazine section you will see articles from five continents that cover a huge range of topics, from basic sciences (such as the pathology of emphysema) to global public health (such as palliative care in developing countries). You will find diverse opinions---for example, on whether President Bush is helping or hindering Africa\'s progress towards tackling HIV, and on whether health professionals should routinely screen women for domestic violence (tell us what you think by taking our poll at [www.plosmedicine.org](www.plosmedicine.org)). And you\'ll find case-based learning materials on meningitis linked to an online video and an online quiz. "The revolutionary idea of anyone being able to read any article is possible." Interpretation of results is an essential part of a medical journal\'s job. Although we expect that many of our readers will be doctors, we hope readers will range from patients wanting to learn about the latest research on their illness, to teachers wanting to use an article in the classroom, to policymakers. Hence, we have several levels of comment on original research. Perspectives, written by an expert, are aimed at readers who are already familiar with the topic. Synopses, written by PLoS Medicine\'s professional editors, should provide any health professional with a quick introduction to an article. Patient summaries provide a starting point for patients to assess the relevance to them of a research paper. We have decided not to be part of the cycle of dependency that has formed between journals and the pharmaceutical industry, an industry that focuses overwhelmingly on the most profitable drugs, thus sidelining many of the world\'s health problems. Medical journals have allowed their interests to become aligned with those of the pharmaceutical industry by printing advertisements for drugs, publishing trials designed by drug companies\' marketing departments, and making profits on reprints used as marketing tools. PLoS Medicine will not accept advertisements for pharmaceutical products or medical devices. Our open-access license allows free distribution of articles, so PLoS cannot benefit from exclusive reprint sales. And we consider as the lowest priority for publication papers that are simply aimed at increasing a drug\'s market share without obvious benefit to patients. We will aim to have the highest levels of transparency in our published papers. We require authors to tell us of any possible competing interests; we in turn will tell readers about them. But, information flow should not be just one-way. Our editorial doors (or at least our E-mail boxes) are always open. We want your feedback on the journal: send us an E-mail or submit an eLetter about any article in the journal, take part in our polls, contribute ideas for the magazine section and submit original research. PLoS Medicine is a journal for the global medical community; we invite you to join in. The editors for PLoS Medicine are Virginia Barbour, James Butcher, Barbara Cohen, and Gavin Yamey. E-mail: <medicine_editors@plos.org>	PubMed Central	2024-06-05T03:55:48.596103	2004-10-19	{ "license": "Creative Commons Zero - Public Domain - https://creativecommons.org/publicdomain/zero/1.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523840/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e22", "authors": [] }
PMC523841	Most patients with celiac disease can eliminate their symptoms---at a price: life-long adherence to a gluten-free diet. This means no wheat, rye, barley, and, until recently, no oats. Then some recent studies suggested that oats did not cause the intestinal inflammation characteristic of the disease, and thus oats are now often included in the celiac disease diet. This is good news for patients coping with severe restrictions on what they can and must not eat, but a study by Ludvig Sollid and colleagues in this issue of PLoS Medicine suggests that oats are not safe in all cases.[](#pmed-0010023-g001){ref-type="fig"} ::: {#pmed-0010023-g001 .fig} ::: {.caption} ###### The celiac diet excludes many cereal products (Photo: National Cancer Institute) ::: ![](pmed.0010023.g001) ::: Like other chronic inflammatory diseases, celiac disease is caused by a complex interplay between genetic and environmental factors, but it is better understood than most. Long believed to be a relatively rare disorder, it is now thought to affect about one in 250 people worldwide. Clinical symptoms are present in less than half of patients and vary considerably. Genetically, almost all patients have one of two predisposing HLA molecules, which determine the context in which their immune system encounters foreign antigens, including gluten proteins found in wheat and other cereals. In individuals with celiac disease, the immune system mounts an abnormal response to gluten, which is characterized by gluten-reactive intestinal T cells and by inflammation and compromised function of the small intestine. Ludvig Sollid and colleagues applied the current understanding of celiac disease and a range of molecular pathology tools to studying the response to oats of nine patients with celiac disease. The nine patients were not a random sample: all of them had been eating oats, and four of them had shown clinical symptoms after oats ingestion. The goal of the study was to characterize the intestinal T cell response to oats in these patients, and to relate it to clinical symptoms and intestinal biopsy results. All patients were on a gluten-free diet and ate oats that were free of contamination by other cereals. Three of the four patients who had reported problems after eating oats showed intestinal inflammation typical of celiac disease, and Sollid and colleagues studied intestinal T cells from these three patients. Two of the five patients who seemed to tolerate oats also had oats-reactive intestinal T cells. Functional study of these T cells showed that they were restricted to celiac-disease-associated HLA molecules and that they recognized two peptides derived from oat avenin that are very similar to peptides of gluten. Taken together, the findings show that intolerance to oats exists at least in some patients with celiac disease, and that those patients have the same molecular reaction to oats that other patients have to wheat, barley, or rye. However, identical reactions were also seen in two of the patients who were clinically tolerant to oats. The authors suggest that these reactions could develop into symptomatic disease after some time delay, but there is no proof that the presence of oats-reactive T cells is an indicator of future symptoms or even of enhanced susceptibility to clinical oats intolerance. Oats are not safe for all patients with celiac disease, but future studies are needed to determine the frequency of oats intolerance.	PubMed Central	2024-06-05T03:55:48.596764	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523841/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e23", "authors": [] }
PMC523842	Acidosis is a major cause of death in patients with malaria, although what causes acidosis is still unclear. One possibility is that hypovolemia contributes to the problem, and that rehydration therapy could be of benefit. Now, Sanjeev Krishna and colleagues have shown that in children with severe malaria dehydration is not severe and is not correlated with other measures of disease severity. "The optimum resuscitation approach in severe childhood malaria remains to be defined," says Nick White (Mahidol University, Thailand), the academic editor of the paper. "The relative advantages of blood, colloids, and crystalloids need to be characterized."[](#pmed-0010024-g001){ref-type="fig"} ::: {#pmed-0010024-g001 .fig} ::: {.caption} ###### Anopheles gambiae, the principal vector of malaria (Photo: Jim Gathany) ::: ![](pmed.0010024.g001) ::: Every year around 200 million people worldwide contract malaria, of whom over a million die. The vast majority of those who die are children under five years, mostly in Africa, since young children have had little chance to acquire any immunity. Fluid resuscitation is generally considered to be a cornerstone of treatment---but how much fluid should be given? Some researchers believe that surrogate signs of fluid depletion---such as tachycardia, reduced capillary refill time, and reduced urine excretion---suggest that there is substantial volume depletion. The reason that the amount of fluid given matters so much is that giving too much, especially of hypotonic solutions, can lead to electrolyte imbalance, especially hyponatremia and hypokalemia. Research efforts have been hampered by not having an easy way to assess in patients the fluid depletion in different compartments of the body, i.e., total body water and extracellular and intracellular water volume. Krishna and colleagues used heavy-water distribution to calculate the total body water and bromide distribution to determine the extracellular volume in 19 children with moderately severe malaria and 16 with severe malaria in Gabon. By subtracting extracellular volume from total body water, they were able to calculate intracellular volume for each child. They also used a less invasive and more rapid method of determining water volumes based on using bioelectrical impedance to calculate the volume. None of the children were severely dehydrated (defined as more than 100 ml/kg depletion), and only three of the children with severe anemia had fluid depletion, which was moderate (60--90 ml/kg depletion). "This challenges the view that dehydration is a major contributor to the pathology of this frequently lethal disease," says White. So based on these data, obtained from a carefully studied, albeit small group of children, what should people who treat children with malaria do? The authors\' first recommendation is that clinicians should think again about how vigorously they rehydrate children, and if they have access to ways of assessing fluid volume more precisely, they should do so (not a trivial undertaking in many hospitals where these children are treated). And certainly the methods used by Krishna and colleagues should undergo wider testing in larger groups of children to confirm their usefulness. Until the worldwide efforts to prevent malaria come to fruition, refining the management of infected children will remain a cornerstone of the efforts against this devastating disease.	PubMed Central	2024-06-05T03:55:48.597387	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523842/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e24", "authors": [] }
PMC523843	T lymphocytes may have an important role in the pathogenesis of smoking-related emphysema, according to a new study by researchers from Houston, Texas, United States. "We now know that T cells are not only present in chronic obstructive pulmonary disease \[COPD\], but are harmful," comments Steven Shapiro from Brigham and Women\'s Hospital, Harvard Medical School, who was not involved in the study. "We also now have a pathway that could be interrupted to prevent lung destruction in COPD." Farrah Kheradmand and colleagues took lung samples from 28 ex-smokers who had been admitted to hospital for lung resection: 18 patients had moderate to severe COPD as well as evidence of emphysema, and ten patients had none. The researchers isolated lung lymphocytes from the samples and used two-color flow cytometry to phenotypically characterize the cells. They found that lymphocytes taken from patients with emphysema expressed more CCR5 and CXCR3 receptors, which are associated with a particular type of T cell called T helper 1 (Th1), than did those from control individuals. By contrast, expression of CCR4 receptors, which are found on T helper 2 (Th2) cells, was very low in both control and emphysema groups.[](#pmed-0010025-g001){ref-type="fig"} ::: {#pmed-0010025-g001 .fig} ::: {.caption} ###### CT image of the lung of subjects with end-stage emphysema next to a photomicrograph of their resected lung stained with H&E ::: ![](pmed.0010025.g001) ::: In a separate experiment, Kheradmand\'s team showed that lung lymphocytes taken from patients with emphysema secreted more of three other proteins---interferon gamma, monokine induced by interferon (MIG), and interferon-inducible protein 10 (IP-10)---than control patients. MIG and IP-10 are known to be produced by injured epithelial cells and are ligands for CXCR3 receptors, which are expressed by Th1 cells. Importantly, the researchers were also able to show that isolated peripheral lung macrophages secreted matrix metalloproteinase-12 (MMP12), an enzyme that degrades elastin---a protein important for lung elasticity---in the lungs, in response to IP-10 and MIG. Together these findings, say the authors, indicate that Th1 cells, but not Th2 cells, are required for producing the elastin-destroying lung environment of emphysema. The researchers now intend to investigate the antigens that drive the Th1-based inflammation that underlies emphysema. "Ultimately, we seek to understand the biochemistry of tobacco smoke that triggers inflammation in the first place, and whether such insight might explain other environmentally triggered lung diseases," explains Kheradmand. "To understand such detailed immune mechanisms, we really need an improved experimental model of disease, and this we are currently working on."	PubMed Central	2024-06-05T03:55:48.597833	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523843/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e25", "authors": [] }
PMC523844	Introduction {#s1} ============ Reliable and comparable analysis of risks to health is an important component of evidence-based policies and programs for disease prevention \[[@pmed-0010027-b01],[@pmed-0010027-b02]\]. An important feature of risk assessment, with implications for specific interventions as well as broad prevention policies, is the distribution of disease burden among population subgroups. These subgroups may be defined by demographic factors, such as age and sex, or by socioeconomic status. Subgroups can also be defined by the level of exposure to a risk factor, if exposures are defined in multiple categories or continuously. Understanding the distribution of risk-factor burden is particularly important for targeting interventions. For example, the large number of road traffic accident injuries and deaths among young adult males may be largely associated with binge alcohol consumption by this group. Interventions that focus on this population subgroup and their specific drinking behaviors may be more effective or cost-effective than, for example, raising alcohol taxes, which would have a more diffuse impact on alcohol consumption. On the other hand, the majority of effects from risk factors such as blood pressure have been found to be among those at moderately elevated levels, motivating interventions beyond those intended for clinical hypertension \[[@pmed-0010027-b03],[@pmed-0010027-b04],[@pmed-0010027-b05]\]. The distributions of the health effects of risk-factor exposure by age and sex or by exposure level have been studied in specific cohorts and for specific risk factors \[[@pmed-0010027-b06],[@pmed-0010027-b07],[@pmed-0010027-b08]\]. Most notably Rose\'s seminal work stated that "a large number of people exposed to a small risk may generate many more cases than a small number exposed to high risk" \[[@pmed-0010027-b09]\]. Using the data from a global and regional assessment of multiple major risk factors, this paper reports the distribution by exposure levels, age, and sex of disease burden attributable to several major risk factors. The findings of this analysis confirm that Rose\'s observations have global relevance and also illustrate important new patterns on specific distributions of disease burden for multiple risks, in different age groups, and in populations at various stages of economic development. Methods {#s2} ======= The methods and data sources for individual risk factors and for estimating population attributable fractions (PAFs) and disease burden attributable to them have been fully described elsewhere \[[@pmed-0010027-b01],[@pmed-0010027-b02]\] and are summarized below. The contribution of a risk factor to disease or mortality relative to some alternative exposure scenario (i.e., PAF, defined as the proportional reduction in population disease or mortality that would occur if exposure to the risk factor were reduced to an alternative exposure scenario \[[@pmed-0010027-b10]\]) is given by the generalized "potential impact fraction": RR(x) is relative risk at exposure level x, P(x) is the population distribution of exposure, P′(x) is the alternative or counterfactual distribution of exposure, and m is the maximum exposure level. The alternative (counterfactual) scenario used in this work is the exposure distribution that would result in the lowest population risk, referred to as the theoretical minimum-risk exposure distribution ([Table 1](#pmed-0010027-t101){ref-type="table"}) \[[@pmed-0010027-b01],[@pmed-0010027-b02]\]. The theoretical minimum exposure distribution was zero in most cases since zero exposure reflected minimum risk (e.g., no smoking). For some risk factors, zero exposure was an inappropriate choice as the theoretical minimum, either because it is physiologically impossible (e.g., body mass index \[BMI\] and cholesterol) or because there are physical lower limits to exposure reduction (e.g., ambient particulate matter concentration and occupational noise). For these risk factors, the lowest levels observed in specific populations and epidemiological studies were used as the theoretical minimum. For factors with protective effects (i.e., fruit and vegetable intake and physical activity), a counterfactual exposure distribution was chosen based on levels in high-intake populations (e.g., fruit and vegetable intake in Greece) and the level to which the benefits may continue given current scientific evidence. Using theoretical minimum exposure distribution as the counterfactual has the advantage of providing a vision of potential gains in population health by risk reduction from all levels of suboptimal exposure in a consistent way across risk factors. ::: {#pmed-0010027-t101 .table-wrap} Table 1 ::: {.caption} ###### Leading Global Risk Factors, Exposure Variables, Theoretical Minima, and Attributable Deaths and Disease Burden (measured in DALYs) in 2000 ::: ![](pmed.0010027.t101) See [Table 1](#pmed-0010027-t101){ref-type="table"} in Ezzati et al. \[[@pmed-0010027-b02]\] for disease outcomes and data sources ^a^ The resulting hemoglobin levels vary across regions and age-sex groups (from 11.66 g/dl in children under five in a region of Southeast Asia to greater than 14.5 g/dl in adult males in developed countries) because the other risks for anemia (e.g., malaria) vary ^b^ Theoretical minimum for alcohol is zero, the global theoretical minimum. Specific subgroups may have a non-zero theoretical minimum ^c^ Theoretical minimum for lead is the blood lead levels expected at background exposure levels. Health effects were quantified for blood lead levels above 5 μg/dl where epidemiological studies have quantified hazards ::: ::: {#pmed-0010027-t102 .table-wrap} Table 1 ::: {.caption} ###### Continued ::: ![](pmed.0010027.t102) ::: ::: {#pmed-0010027-t103 .table-wrap} Table 1 ::: {.caption} ###### Continued ::: ![](pmed.0010027.t103) ::: Estimates were made for eight age groups (0--4, 5--14, 15--24, 25--44, 45--59, 60--69, 70--79, and 80+ y), both sexes, and 14 Global Burden of Disease subregions ([Table 2](#pmed-0010027-t002){ref-type="table"}). PAFs were estimated for mortality and incidence and were applied to regional cause-specific mortality and disease burden from the World Health Organization (WHO) Global Burden of Disease database ([Table 1](#pmed-0010027-t101){ref-type="table"}). Burden of disease, reported annually in the annexes of the World Health Report, was expressed in disability-adjusted life-years (DALYs) with methods and assumptions described elsewhere \[[@pmed-0010027-b11]\]. Aggregate results (both mortality and disease burden) for all ages, sexes, and exposure levels have been reported elsewhere \[[@pmed-0010027-b01],[@pmed-0010027-b02]\]. Many risks act simultaneously to cause disease, and joint effects have also been estimated \[[@pmed-0010027-b12]\], though the separate effects are presented in this paper. The aim of the analyses reported here was to obtain estimates of the distribution of disease burden by age, sex, and exposure level. To make separate estimates of disease burden by exposure level, the relationship in [equation 1](#pmed-0010027-e001){ref-type="disp-formula"} was re-estimated with the entire exposure distribution divided into "narrow bands," with each band corresponding to one level of exposure, and the estimation repeated for each such exposure band. ::: {#pmed-0010027-t002 .table-wrap} Table 2 ::: {.caption} ###### Global Burden of Disease 2000 Subregions ::: ![](pmed.0010027.t002) High-mortality developing regions: AFR-D, AFR-E, AMR-D, EMR-D, and SEAR-D. Lower-mortality developing regions: AMR-B, EMR-B, SEAR-B, and WPR-B. Developed regions: AMR-A, EUR-A, EUR-B, EUR-C, and WPR-A ^a^ A, very low child mortality and very low adult mortality; B, low child mortality and low adult mortality; C, low child mortality and high adult mortality; D, high child mortality and high adult mortality; E, high child mortality and very high adult mortality ::: Results {#s3} ======= The distribution of deaths and DALYs attributable to the risk factors by age and sex is shown in [Table 3](#pmed-0010027-t003){ref-type="table"}. Disease burden attributable to being underweight and to micronutrient deficiencies in children was equally distributed among males and females. The total all-age disease burden from iron and vitamin A deficiencies was greater in females because of effects on maternal mortality and morbidity conditions. The disease burden of other diet-related risks, physical inactivity, environmental risks, and unsafe sex (sex with an infected partner without any measures to prevent infection) occurred almost equally in males and females. However, approximately 80% of disease burden from addictive substances and 60%--90% from occupational risks occured among men (bearing in mind that the assessment considered only formal employment). Women experienced an estimated two-thirds of the disease burden from childhood sexual abuse and all of the burden caused by a lack of contraception. ::: {#pmed-0010027-t003 .table-wrap} Table 3 ::: {.caption} ###### Distribution of Risk-Factor-Attributable Deaths and Disease Burden (DALYs) by Age and Sex ::: ![](pmed.0010027.t003) See [Table 1](#pmed-0010027-t101){ref-type="table"}, and Figure 1 in \[[@pmed-0010027-b02]\], for definition of each risk factor, data sources and methods, and total magnitude of mortality and DALYs ::: The estimated disease burden from childhood and maternal undernutrition; unsafe water, sanitation, and hygiene; and global climate change (much of whose estimated effects are mediated through nutritional and water variables) was almost exclusively among children under 5 y of age. For these risks, more than 85% of total attributable burden occurred in this age group, with the exception of iron deficiency, for which 30% of burden was borne by women of childbearing age. Only a small fraction of disease burden from the risk factors considered was among the 5--14 y olds. This was because some of the leading diseases of this age group (e.g., injuries and depression) have complex causes that could not easily be included in the current risk-based framework \[[@pmed-0010027-b12]\]. For other leading diseases of this group (e.g., diarrhea and lower respiratory infections), most epidemiological studies have focused on children under 5 y of age and do not provide hazard estimates for older children. The disease burden from other diet-related risks, tobacco, and occupational risks (except injuries and back pain) was almost equally distributed among adults above and below the age of 60 y. For example, 43% of disease burden due to blood pressure and 61% of burden due to tobacco occurred in the 15--59-y age group. More than 90% of disease burden attributable to lack of contraception, illicit drugs, occupational ergonomic stressors, risk factors for injury, and childhood sexual abuse occurred in adults below the age of 60 y. Similarly, 67%--87% of the disease burden attributable to alcohol, unsafe sex, and unsafe health-care injections arose from events occurring between 15 and 59 y of age. Most of the risks whose hazards are concentrated in the younger adults are causes of injuries, neuropsychiatric diseases, maternal conditions, and HIV/AIDS. Assessment by the level of economic and demographic development illustrated that, with the exception of alcohol, which has global presence, the majority of disease burden from risks for mortality and disease among young adults is concentrated in developing countries (see [Figure 1](#pmed-0010027-g001){ref-type="fig"} in \[[@pmed-0010027-b02]\]). Stratification of economic and demographic development was also a determinant of the age distribution patterns for risk factors for chronic diseases, which occurred in younger age groups in developing countries than in developed regions. For example, in high-mortality developing regions, 69% of the tobacco burden occurred in people aged 15--59 y, whereas this share was 63% for lower-mortality developing countries and 55% for developed countries. ::: {#pmed-0010027-g001 .fig} Figure 1 ::: {.caption} ###### Distribution by Exposure Level of Attributable Disease Burden Due to Selected Continuous Risk Factors Figure 1 shows the distribution of the estimated cardiovascular disease (CVD) burden of disease (in DALYs) attributable to four major continuous risk factors, by exposure levels. Half the attributable burden occurs to the left of the solid vertical line and half occurs to the right. The dashed vertical lines indicate commonly used thresholds---150 mm Hg for hypertension, 6.0 mmol/l for hypercholesterolemia, and 30 kg/m^2^ for obesity. The blood pressure and cholesterol levels plotted are the estimated usual levels \[[@pmed-0010027-b22]\], which tend to have a smaller SD than levels based on one-off measurements commonly used in population surveys, because of normal day-to-day and week-to-week fluctuations. For example, the distribution of usual blood pressure is about half as wide as the distribution of one-off blood pressure measures, and so many fewer people would be classified as hypertensive (or hypotensive) if classifications were based on usual rather than one-off blood pressure. Thus, if a population mean SBP was 134 mm Hg, the SD of once-only measures might be 17 mm Hg (with about 18% of the population having one-off SBP over 150 mm Hg), and the SD of usual SBP based on many measures would be about 9 mm Hg (hence about 5% of the population would have usual SBP over 150 mm Hg). ::: ![](pmed.0010027.g001) ::: The distributions of attributable risk-factor burden by exposure levels are shown in [Table 4](#pmed-0010027-t004){ref-type="table"} for those risks quantified using categorical variables and in [Figure 1](#pmed-0010027-g001){ref-type="fig"} for those with continuous variables. For most of these risks a substantial proportion of attributable disease burden occurred among those with modest elevations of risk. For example, while 35% of the large disease burden from being underweight in childhood, the leading risk factor for global loss of healthy life, occurred in severely underweight children who would be subject to clinical interventions (i.e., more than three standard deviations \[SDs\] from referent group median), the rest occurred in children only 1--3 SDs below the median. Even though the relative risks for the latter group are much lower, the number of children exposed to risk at this level is so great that the total attributable disease burden amounted to much more than that occurring in the severe category. Similarly, 52% of the attributable burden from physical inactivity occurred among those engaged in some but less than the recommended level of 2.5 h per week of moderate-intensity activity. For unsafe water, sanitation, and hygiene, almost all of the attributable disease burden was distributed among three of the five at-risk exposure categories, with approximately equal levels. This reflects the fact that the exposure categories were defined as the presence or absence of technology-based water and sanitation interventions. During decades of water and sanitation projects, many countries have "clustered" in a limited number of technology groups. There is likely to be large heterogeneity of exposure within each exposure category, however, because of factors such as quantity of water consumed and hygiene behavior \[[@pmed-0010027-b13]\]. ::: {#pmed-0010027-t004 .table-wrap} Table 4 ::: {.caption} ###### Distribution by Exposure Level of Attributable Burden Due to Selected Categorical Risk Factors ::: ![](pmed.0010027.t004) ::: [Figure 1](#pmed-0010027-g001){ref-type="fig"} shows that at the aggregated level a substantial proportion of the attributable disease burden for high blood pressure, cholesterol, and BMI and low fruit and vegetable intake occurred in the "mid-range" exposures. For example, the second and third quartiles (i.e., half of attributable disease burden) occurred at the following levels: systolic blood pressure (SBP) of 130--150 mm Hg, cholesterol of 5.0--6.1 mmol/l, BMI of 25--32 kg/m^2^, and fruit and vegetable intake of 150--300 g/d (2--4 servings/d). This was similar to or greater than the amount of disease burden occuring among individuals with risk-factor levels above the commonly used, but arbitrary, thresholds for hypertension, hypercholesterolemia, and obesity indicated in [Figure 1](#pmed-0010027-g001){ref-type="fig"}. Despite the above finding on the important role of moderately elevated levels in total disease burden, the actual patterns of how disease burden is distributed among exposure levels varied across different regions and risk factors ([Figure 2](#pmed-0010027-g002){ref-type="fig"}). For example, [Figure 2](#pmed-0010027-g002){ref-type="fig"} shows that a comparatively larger fraction of the disease burden attributable to elevated blood pressure, cholesterol, and BMI occurred at lower levels in developing regions compared to developed regions, mainly because of lower age-specific exposure levels in those populations. [Figure 2](#pmed-0010027-g002){ref-type="fig"} also shows that the skewness of the distribution of disease burden was not substantially different across different age groups for BMI. This is because the comparatively larger relative risk per unit BMI at younger ages (which leads to more right-hand skew) is counterbalanced by the comparatively lower BMI at younger ages (which leads to left-hand skew). Therefore, the different distributions of BMI-attributable disease burden by region resulted not from the different age structures of populations across major world regions, but rather from the lower age-specific BMI levels in those countries \[[@pmed-0010027-b14]\]. This is in contrast to blood pressure, for which disease burden in younger age groups occurred at lower exposures because the age patterns of exposure and relative risk do not entirely compensate. ::: {#pmed-0010027-g002 .fig} Figure 2 ::: {.caption} ###### Distribution of Attributable Cardiovascular Disease Burden Due to BMI, Blood Pressure, and Cholesterol by Exposure Level, Age, and Level of Development Conventions as for [Figure 1](#pmed-0010027-g001){ref-type="fig"}. ::: ![](pmed.0010027.g002) ::: Discussion {#s4} ========== The findings reported here should be considered within the context of limited available data and are subject to uncertainty, which varies across risk factors and is generally most marked in developing countries. Full discussion of uncertainty in the basic data sources and parameters is provided elsewhere and includes the uncertainties in estimates of disease incidence, duration, severity, and disability weighting \[[@pmed-0010027-b01],[@pmed-0010027-b02]\]. Uncertainty in this risk assessment is by far dominated by absence or limitations of direct studies on exposure, hazard, and background disease burden, rather than parameter uncertainty, such as sampling error. This has motivated innovative assumptions and extrapolations even in the case of the best-studied risk factors in a limited number of countries \[[@pmed-0010027-b06]\]. While estimates of hazard size in individual studies were as much as possible adjusted for confounding effects, it is likely that residual confounding effects remain to some extent, hence leading to errors in estimation. Extrapolation of hazard from a limited number of studies to other populations is another source of potential error. While the robustness of relative measures of risk has been confirmed for more proximal factors in studies across populations \[[@pmed-0010027-b15],[@pmed-0010027-b16],[@pmed-0010027-b17]\], their extrapolation is an important source of uncertainty for more distal risks (e.g., childhood sexual abuse) or those whose effects are heterogeneous (e.g., alcohol and injuries versus alcohol and cancer). Direct exposure data for many risk factors are limited because of both difficulties in their measurement and underinvestment in risk-factor surveillance, especially in developing countries. Of particular relevance to the current analysis is the fact that, due to data limitations, some risks were measured using categorical variables (e.g., indoor smoke from solid fuels, underweight, and physical inactivity) even though the health effects occur along a continuum. Other risk factors were represented using indirect or aggregate indicators (e.g., smoking impact ratio for accumulated hazards of smoking, and motor vehicle accident registries for alcohol-caused accidents) that do not allow quantification of risks along continua of exposure. Two important sources of uncertainty with implications for interventions are correlations among multiple risk factors and the skewness of risk-factor distribution. Because risks are often correlated (e.g., undernutrition, poor water and sanitation, and the use of solid fuels are more common among poor households in developing countries, and smokers are more likely to have poor diets), the contributions of high-exposure groups are likely to be underestimated. In addition to being a source of underestimation at higher exposure levels, risk-factor correlation implies that the same individuals and groups are at the high end of multiple risk factors. Positive (rightward) skewness of exposure distribution, not modeled here, would also lead to an underestimation of events occurring in those who were hypertensive, hypercholesterolemic, or obese. The importance of skewness is emphasized by the observation that the recent increase in the proportion of people who are overweight and obese (e.g., in the United States) has involved not only a shift in the distribution, but also increasing skewness, which has extended the high-exposure tail. Coupled with risk-factor correlation, this should motivate more systematic data collection and reporting to determine mean exposure as well as the complete shape of distribution. The temporal nature of risk-factor exposure also has implications for the current cross-sectional estimates. There is an expectation that the size and rank order of risk-factor burden will alter in coming decades as a result of changes in exposure levels and delays between exposure and hazard. For example, it is predicted that by 2020 the leading risks to health will be (1) unsafe sex, principally because of HIV/AIDS and driven by increasing exposure, and (2) tobacco, because of market expansion of tobacco products in developing countries and delayed temporal effects on major health outcomes such as lung cancer \[[@pmed-0010027-b01]\]. This analysis in multiple age and exposure categories, or along a continuum of exposures, showed that, globally, a considerable proportion of the disease burden attributable to major risk factors occurred among those with only moderately raised levels, not the extremes such as those that define hypertension, obesity, or severe malnutrition. Further for many chronic-disease risk factors, such as tobacco and high blood pressure, as well as risk factors for injuries and sexual and reproductive health, significant proportions of disease burden occurred from events in middle ages, especially in developing countries. The concentration of disease burden from such a large number of risk factors in those below 60 years of age illustrates the large, and at times neglected, disease burden from risks that affect young adults in developing nations, with important consequences for economic development \[[@pmed-0010027-b18]\]. The distribution of risks and their determinants in a population have major implications for strategies of prevention. As stated by Rose, risk typically increases across the spectrum of a risk factor \[[@pmed-0010027-b08]\]. Rose\'s work led to one of the most fundamental axioms in disease prevention across risk factors: "A large number of people exposed to a small risk may generate many more cases than a small number exposed to high risk." The analysis in this work showed that the risk factors for many of the major global diseases, such as lower respiratory infections, diarrhea, ischemic heart disease, and stroke exhibit such behavior, because they are caused by risks that occur along a continuum. Therefore, managing individual, high-risk crises, while appropriate for individuals, can only have a limited preventive effect at the population level and over long time periods because it relies on having adequate discriminative ability to predict future disease, and requires continued and expensive screening for new high-risk individuals. In contrast, population-based strategies that seek to shift the whole distribution of risk factors have the potential to substantially reduce total disease burden, and possibly over long time periods, if the interventions alter the underlying risk behaviors or their socioeconomic causes. This is particularly relevant in the context of risk factors such as those related to under- or overnutrition that affect societies in all stages of development. For example, a policy to reduce salt content in manufactured foods would result in a leftward shift in the population distribution of blood pressures and a surprisingly large reduction in cardiovascular disease \[[@pmed-0010027-b05]\]. Another example would be population-level measures that affected energy intake (such as the availability and price of energy-dense foods) and/or expenditure (such as the level of motorization and mechanization)---these can be expected to determine the distribution of a population\'s BMI levels, and hence largely determine its level of type 2 diabetes \[[@pmed-0010027-b01]\]. There were distinct patterns for the distribution of disease burden across risk factors, and across regions at various stages of development. At the extreme of these diverse patterns, for risk factors with acute exposure and acute outcomes, the underlying relationship is considerably more complex. For example while in many societies most alcohol-attributable injury (e.g., traffic accidents) involves people who on average drink moderate amounts \[[@pmed-0010027-b19]\], these people would be on the more extreme, high-risk spectrum in a different dimension: volume of drinking right before the injury. Therefore, if exposure to risk factors is clustered or the risk relationship does not follow a linear pattern, high-exposure groups may indeed play a disproportionately important role \[[@pmed-0010027-b20],[@pmed-0010027-b21]\]. In summary, the analysis presented in this paper confirms and extends to a global level previous work indicating that the impact of many major risks is important across their exposure levels, not just among people with particularly high levels \[[@pmed-0010027-b08]\]. This analysis also illustrates that, beyond the general principle of population-wide shifts, there are important population-specific and risk-factor-specific patterns in how the burden of disease attributable to risk factors is distributed. Systematic assessment of multiple risks within any given population can provide the basis for selecting packages of interventions that include population-wide measures as well as highly targeted interventions provided to much smaller subsections of the population with constellations of major risks \[[@pmed-0010027-b01],[@pmed-0010027-b05],[@pmed-0010027-b18]\]. ::: {.caption} ###### Patient Summary ::: {#sb1a} ### Background {#sb1a1} Health policy makers must understand existing health risks and which diseases cause the most health problems. The Global Burden of Disease database, maintained by the World Health Organization, collects information from around the world on risk factors such as malnutrition, childbirth, tobacco use, and cholesterol levels, as well as on diseases such as depression, blindness, and diarrhea. This information can be used to target health interventions. ### What Did the Researchers Do and Find? {#sb1a2} These researchers determined for 26 major risk factors the distribution of disease burden by age, sex, and exposure level. They found that many risk factors (such as high blood pressure and high cholesterol levels) occur across populations, and are not confined to one sex or age group. And for most risk factors, exposure to moderate risks (which is usually more common than exposure to severe risk) is responsible for causing most disease. ### What Do the Results Mean, and Who Will Use Them? {#sb1a3} Measures that reduce exposure to risk factors across whole populations, if only by a modest extent, can achieve large reductions in disease burden. This information is important for people who set global health policies and priorities. ### What Are the Problems with the Study? {#sb1a4} There are many challenges in estimating the impact of different major risks, each of which has distinct effects on a number of diseases. In addition, exposure to some risk factors today will cause disease only many years from now, so the picture will change over time. The major finding of the global distribution of many major risks is secure, but the exact extent of this remains uncertain due to the paucity of data in developing countries. This work was sponsored by the National Institute of Aging grant PO1-AG17625. The sponsor had no influence on analysis and results. Anthony Rodgers holds a National Heart Foundation Senior Fellowship. We thank Clarissa Gould-Thorpe for manuscript preparation. We thank Mie Inoue, Doris Ma Fat, Susan Piccolo, Chalapati Rao, Kenji Shibuya, Niels Tomijima, and Marie-Claude von Rulach for assistance with global burden of disease databases and project management. Comparative Risk Assessment Collaborating Group =============================================== Core, methodology, statistical analysis, editorial and peer review, and writing: Majid Ezzati (Harvard University), Anthony Rodgers (University of Auckland), Alan D. Lopez (University of Queensland), Stephen Vander Hoorn (University of Auckland), Christopher J. L.Murray (Harvard University). Childhood and maternal underweight: Steven Fishman (Johns Hopkins University), Laura E.Caulfield (Johns Hopkins University), Mercedes de Onis (WHO), Monika Blössner (WHO), Adnan A. Hyder (Johns Hopkins University), Luke Mullany (Johns Hopkins University), Robert E. Black (Johns Hopkins University). Iron deficiency anemia: Rebecca J. Stoltzfus (Cornell University), Luke Mullany (Johns Hopkins University), Robert E. Black (Johns Hopkins University). Vitamin A deficiency: Amy J. Rice (Johns Hopkins University), Keith P, West, Jr. (Johns Hopkins University), Robert E. Black (Johns Hopkins University). Zinc deficiency: Laura E. Caulfield (Johns Hopkins University), Robert E. Black (Johns Hopkins University). High blood pressure: Carlene Lawes (University of Auckland), Stephen Vander Hoorn (University of Auckland), Malcolm Law (St Bartholomew\'s and Royal London School of Medicine), Paul Elliott (Imperial College School of Medicine), Stephen MacMahon (University of Sydney), Anthony Rodgers (University of Auckland). High cholesterol: Carlene Lawes (University of Auckland), Stephen Vander Hoorn (University of Auckland), Malcolm Law (St Bartholomew\'s and Royal London School of Medicine), Stephen MacMahon (University of Sydney), Anthony Rodgers (University of Auckland). Overweight and obesity (high BMI): W. Philip T. James (International Obesity Task Force), Rachel Jackson-Leach (International Obesity Task Force), Cliona Ni Mhurchu (University of Auckland), Eleni Kalamara (International Obesity Task Force), Maryam Shayeghi (International Obesity Task Force), Neville J. Rigby (International Obesity Task Force), Chizuru Nishida (WHO), Anthony Rodgers (University of Auckland). Low fruit and vegetable consumption: Karen Lock (London School of Hygiene and Tropical Medicine), Joceline Pomerleau (London School of Hygiene and Tropical Medicine), Louise Causer (London School of Hygiene and Tropical Medicine), Martin McKee (London School of Hygiene and Tropical Medicine). Physical inactivity: Fiona C. Bull (Loughborough University), Tim Armstrong (WHO), Tracy Dixon (Australian Institute of Health and Welfare), Sandra Ham (United States Centers for Disease Control and Prevention), Andrea Neiman (United States Centers for Disease Control and Prevention), Michael Pratt (United States Centers for Disease Control and Prevention). Tobacco: Majid Ezzati (Harvard University), Alan D. Lopez (University of Queensland). Alcohol: Jürgen Rehm (Addiction Research Institute and University of Toronto), Robin Room (Stockholm University), Maristela Monteiro (WHO), Gerhard Gmel (Swiss Institute for the Prevention of Alcohol and Other Drug Problems), Kathryn Graham (University of Western Ontario), Nina Rehn (WHO), Christopher T. Sempos (University of Buffalo), Ulrich Frick (Psychiatric University Hospital Zurich), David Jernigan (Georgetown University). Illicit drugs: Louisa Degenhardt (University of New South Wales), Wayne Hall (University of Queensland), Matthew Warner-Smith (University of New South Wales), Michael Lynskey (University of New South Wales). Unsafe sex: Emma Slaymaker (London School of Hygiene and Tropical Medicine), Neff Walker (Joint United Nations Programme on HIV/AIDS), Basia Zaba (London School of Hygiene and Tropical Medicine), Martine Collumbien (London School of Hygiene and Tropical Medicine). Non-use and use of ineffective methods of contraception: Martine Collumbien (London School of Hygiene and Tropical Medicine), Makeda Gerressu (London School of Hygiene and Tropical Medicine), John Cleland (London School of Hygiene and Tropical Medicine). Unsafe water, sanitation, and hygiene: Annette Prüss-Ustun (WHO), David Kay (University of Wales), Lorna Fewtrell (University of Wales), Jamie Bartram (WHO). Urban air pollution: Aaron Cohen (Health Effects Institute), Ross Anderson (St George\'s Hospital Medical School and University of London), Bart Ostro (California Environmental Protection Agency), Kiran Dev Pandey (World Bank), Michal Krzyzanowski (WHO), Nino Künzli (University of Southern California), Kersten Gutschmidt (WHO), Arden Pope (Brigham Young University), Isabelle Romieu (Instituto Nacional de Salud Publica de Mexico), Jonathan Samet (Johns Hopkins University), Kirk Smith (University of California at Berkeley). Indoor smoke from household use of solid fuels: Kirk R. Smith (University of California at Berkeley), Sumi Mehta (Health Effects Institute), Mirjam Feuz (Federal Office of Public Health of Switzerland). Lead: Annette Prüss-Ustun (WHO), Lorna Fewtrell (University of Wales), Philip Landrigan (Mount Sinai School of Medicine), José Luis Ayuso (Universidad Autonoma de Madrid). Global climate change: Anthony McMichael (Australian National University), Diarmid Campbell-Lendrum (WHO), Sari Kovats (London School of Hygiene and Tropical Medicine), Sally Edwards (London School of Hygiene and Tropical Medicine), Paul Wilkinson (London School of Hygiene and Tropical Medicine), Frank Tanser (Medical Research Council of South Africa), David Le Sueur (deceased) (Medical Research Council of South Africa), Michael Schlesinger (University of Illinois at Urbana-Champaign), Natasha Andronova (University of Illinois at Urbana-Champaign), Robert Nicholls (University of Middlesex), Theresa Wilson (University of Middlesex), Simon Hales (University of Otago). Occupational risk factors for injuries: Marisol Concha (Associacion Chilena de Seguridad of Chile), Deborah Imel Nelson (WHO), Marilyn Fingerhut (WHO), Annette Prüss-Ustun (WHO), James Leigh (University of Sydney), Carlos Corvalan (WHO). Occupational carcinogens: Tim Driscoll (University of Sydney), Deborah Imel Nelson (WHO), N. Kyle Steenland (United States National Institute for Occupational Safety and Health), James Leigh (University of Sydney), Marisol Concha (Associacion Chilena de Seguridad), Annette Prüss-Ustun (WHO), Marilyn Fingerhut (WHO), Carlos Corvalan (WHO). Occupational airborne particulates: Tim Driscoll (University of Sydney), N. Kyle Steenland (United States National Institute for Occupational Safety and Health), Deborah Imel Nelson (WHO), James Leigh (University of Sydney), Marisol Concha (Associacion Chilena de Seguridad), Marilyn Fingerhut (WHO). Occupational ergonomic stressors: Annette Prüss-Ustun (WHO), Laura Punnett (University of Massachusetts at Lowell), SangWoo Tak (University of Massachusetts at Lowell), Deborah Imel Nelson (WHO), Marilyn Fingerhut (WHO), James Leigh (University of Sydney), Sharonne Phillips (Occupational Ergonomics). Occupational noise: Deborah Imel Nelson (WHO), Marisol Concha (Associacion Chilena de Seguridad), Marilyn Fingerhut (WHO). Contaminated injections in health-care settings: Anja M. Hauri (Staatliches Untersuchungsamt Hessen), Gregory L. Armstrong (United States Centers for Disease Control and Prevention), Yvan J. F. Hutin (WHO). Child sexual abuse: Gavin Andrews (University of New South Wales), Justine Corry (University of New South Wales), Cathy Issakidis (University of New South Wales), Tim Slade (University of New South Wales), Heather Swanston (University of New South Wales). Poverty and risk: Tony Blakely (University of Otago), Simon Hales (University of Otago), Charlotte Kieft (University of Otago), Nick Wilson (University of Otago), Alistair Woodward (University of Auckland). Citation: Rodgers A, Ezzati M, Vander Hoorn S, Lopez AD, Lin RB, et al. (2004) Distribution of major health risks: Findings from the Global Burden of Disease study. PLoS Med 1(1): e27. BMI : body mass index DALY : disability-adjusted life-year PAF : population attributable fraction SBP : systolic blood pressure SD : standard deviation WHO : World Health Organization [^1]: Competing Interests: The authors have declared that no competing interests exist. ADL is a member of the editorial board of PLoS Medicine. [^2]: Author Contributions: AR, ME, ADL and CJLM designed the study. SVH and RBL analyzed the data. AR, ME, SVH, and ADL contributed to writing the paper.	PubMed Central	2024-06-05T03:55:48.598276	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523844/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e27", "authors": [ { "first": "Anthony", "last": "Rodgers" }, { "first": "Majid", "last": "Ezzati" }, { "first": "Stephen", "last": "Vander Hoorn" }, { "first": "Alan D", "last": "Lopez" }, { "first": "Ruey-Bin", "last": "Lin" }, { "first": "Christopher J. L", "last": "Murray" } ] }
PMC523845	There is a much-quoted saying, attributed to the epidemiologist Geoffrey Rose: "A large number of people exposed to a small risk may generate many more cases than a small number exposed to a very high risk." This is true for many individual risk factors such as salt intake (linked to high blood pressure and cardiovascular disease) and speeding on the highway (linked to injuries and accidents). Does it apply to many other global health risks? The study by Anthony Rodgers and colleagues suggests that it does. To develop effective health policies, one must understand the existing health risks and disease burdens. On a worldwide scale, this is a tough challenge. The Global Burden of Disease Database, maintained by the World Health Organization (WHO), collects data from countries around the world on risk factors such as tobacco, malnutrition, childhood abuse, unsafe sex, childbirth, and cholesterol levels, as well as on disease burdens, for example depression, blindness, and diarrhea. A large group of scientists from all over the world has developed a framework to analyze these data. To compare different risks or burdens, they calculate disability-adjusted life-years, or DALYs---the number of healthy life years lost because of a particular disease or risk factor.[](#pmed-0010029-g001){ref-type="fig"} ::: {#pmed-0010029-g001 .fig} ::: {.caption} ###### Tobacco is a major player in the global burden of disease (Photo: Bill Branson) ::: ![](pmed.0010029.g001) ::: Rodgers and colleagues used data from the WHO database for 26 risk factors and from 14 epidemiological subregions of the world to calculate the proportion of risk-factor-attributable disease burden in different population subgroups defined by age, sex, and exposure level. For being underweight in childhood, for example---the leading risk factor for global loss of healthy life---they found that only 35% of the disease burden occurred in severely underweight children, the rest occurred in those only moderately underweight. The relative risks for the moderately underweight are much lower, but the number of children in that category is so large that the total attributable burden amounted to almost two-thirds of the total global burden of disease for that risk factor. The analysis confirms---and extends to a global level---previous research showing that many major health risks are important across the range of exposure levels, not just among individuals exposed to high levels of risk. It also points to risk factors that are particularly prevalent among specific populations and age groups, and for which highly targeted interventions could be effective. Despite numerous caveats and limitations of studies like this one, such analyses are essential aids in guiding the distribution of limited funds to lower the burden of life years lost to premature death and disability.	PubMed Central	2024-06-05T03:55:48.600992	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523845/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e29", "authors": [] }
PMC523846	Nevirapine and efavirenz are the most commonly prescribed of the class of antiretroviral drugs called non-nucleoside reverse transcriptase inhibitors (NNRTIs). Efavirenz has the advantage of once-daily dosing. In a recent study called the 2NN study (Lancet 363: 1253--1263), it appeared to be only marginally superior to nevirapine in terms of clinical success and virological suppression. Van Leth and colleagues have now shown that while nevirapine and efavirenz both raise high-density lipoprotein (HDL) cholesterol (the "good" type of cholesterol), the overall lipid profile is better with nevirapine than with efavirenz. "These data suggest that nevirapine may be preferable to efavirenz in HIV-infected adults with other cardiovascular risk factors," says the study\'s academic editor, Andrew Carr of St. Vincent\'s Hospital in Darlinghurst, Australia. "However, perceived cardiovascular risk is only one factor that would affect the choice between these two drugs." Van Leth and colleagues prospectively analyzed the lipids of patients enrolled in the 2NN study, a randomized, open-label efficacy study that included adults with HIV who had never been on antiretroviral drugs. All patients were given stavudine and lamivudine and were then randomized into three treatment groups: nevirapine, efavirenz, or both. For the lipid analysis, which was preplanned, the researchers included only the nevirapine and efavirenz groups (417 and 289 patients, respectively). This was because the 2NN study showed that simultaneous use of nevirapine and efavirenz should be avoided---the combination is associated with increased toxicity without increased efficacy. The increase in HDL cholesterol was significantly higher with nevirapine than with efavirenz. There was a decrease in the ratio of total cholesterol to HDL cholesterol with nevirapine and an increase with efavirenz. The study does not prove, however, that the rise in HDL cholesterol seen with NNRTIs (especially nevirapine) actually leads to a reduction in coronary heart disease. "There are no vascular functional data," says Carr, "or clinical vascular endpoint data that confirm that the statistically significant lipid differences observed are clinically significant." The study was funded by Boehringer Ingelheim, the manufacturer of nevirapine. The authors clearly state that the company had "a nonbinding input on issues of study design and analyses" but it had "no influence on reporting of the data or the decision to publish." Despite its limitations, van Leth and colleagues\' study "moves clinicians and patients away from 'one-size-fits-all' antiretroviral therapy," says Carr. "It takes us further along the path of choice of antiretroviral therapy being individualized according to other patient comorbidities and risk factors, as well as therapy simplicity and side effects."	PubMed Central	2024-06-05T03:55:48.601362	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523846/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e30", "authors": [] }
PMC523847	The Internet is awash with medical information. Eight hundred million people have direct access to the Internet \[[@pmed-0010031-b1]\], and in the United States over 60% have searched for health or medical information on the Web \[[@pmed-0010031-b2]\]. Go to any search engine and type in the name of a disease or drug, and you will be directed to hundreds of sites, ranging from the sound and useful to the quackish and dangerous. Google "medical" and you get 85 million pages, "drug," 40 million, and "health," 230 million. But something is conspicuously missing. The most reliable medical information on the Internet---the contents of peer-reviewed medical journals---is hidden from the public and most of the world\'s physicians. Although most medical journals are available online, their publishers limit access to those who choose, and can afford, to pay for access. This should not, and need not, be so. In the 19th and early 20th centuries independent physicians and small medical societies, interested in making the best new medical knowledge available to doctors, students, and the public, began to publish general medical journals containing case reports, ideas for new treatments, and the results of medical experiments. These pioneers took advantage of the best available technology for disseminating information, printing titles like The Lancet, The New England Journal of Medicine, and The Journal of the American Medical Association on cheap paper and selling them to subscribers at a few pennies a copy. For more than a century, printed journals like these were the dominant means of conveying medical knowledge around the world. But technology has changed. The Internet is now the most economical and efficient conduit for the delivery of information to most places. Publishers of medical journals realize this---when the Internet took off, they took their journals online. But while they adapted their means of distribution to the 21st century, they left their business model in the 19th century, continuing to charge readers for access just as they had done for their printed journals. This has been good for business---medical publishing has never been more profitable---but it comes at a huge cost. The established medical publishers have turned their back on the opportunity to make the latest and best medical information available to anyone with an Internet connection. With the launch of PLoS Medicine, we are embracing this opportunity. "Everything we publish is immediately, freely available online throughout the world, with no restrictions on distribution, copying, printing, or legitimate use." Everything published in PLoS Medicine is immediately freely available online throughout the world, with no restrictions on distribution, copying, printing, or legitimate use. Of course, it costs us money to publish this journal, and we must cover our expenses. But the fee-for-access business model that made perfect sense for the printed journal is no longer consistent with the mission of medical publishing because it needlessly limits the reach of the medical literature. And so we have adopted a new model. Instead of charging readers for access to our journal, we ask the authors of accepted research articles to pay a publication fee to cover the costs of peer review, editorial oversight, and production. This "open access" business model ensures our financial health as a publisher while allowing us to convey everything we publish to the widest possible audience. Of course, we do not expect authors to cover publication costs personally---rather, we expect the government agencies, companies, foundations, research institutions, hospitals, or universities that sponsor the research to pay the fee. These organizations have always considered the wide dissemination of the results of the research they support to be an integral part of their mission. Virtually every leading sponsor of medical research has announced its willingness to pay for open-access publication, the costs of which average less than one percent of the cost of the research itself---a small price to pay to ensure that everyone who could benefit from their research can benefit from it. We realize that not everyone with something important to convey in a medical journal has access to such funds. To ensure that we don\'t replace a barrier to access with barriers to publication, we\'ve raised money to cover the publication costs of articles whose authors are unable to pay them. And, for every PLoS journal, an author\'s ability to pay will never be a consideration in our decision to publish an article. Despite its obvious benefits, open-access publication has met with fierce opposition. Established medical publishers---now businesses more than forces for change---see open-access not as an opportunity to fulfill a mission of public service but as a threat to their lucrative businesses. They contend that their journals still serve the community well, and object that open access threatens their very existence. This is nonsense! "It is our responsibility as publishers and members of the medical community not only to give patients access to the medical literature, but to provide them with tools to use it wisely." The Wellcome Trust, the world\'s largest charitable sponsor of biomedical research, seeking to ensure that the results of the science it funds are "disseminated widely and freely available to all," recently commissioned a thorough analysis of the scientific and medical publishing industry \[[@pmed-0010031-b3]\]. It concluded that the current market "does not operate in the long-term interest of the research community," and issued a strong statement in support of open access \[[@pmed-0010031-b4]\]. Responding to concerns about journal finances, the trust commissioned a detailed economic analysis of open-access publishing \[[@pmed-0010031-b5]\], based on which it concluded that "the open access model of scientific publishing---where the author of a research paper pays for peer reviewed research to be made available on the web free to all who wish to use it---is economically viable, guarantees high quality research and is a sustainable option which could revolutionise the world of traditional scientific publishing" \[[@pmed-0010031-b6]\]. (This report, freely available online, is an excellent resource for anyone with questions about the economics of open-access publishing). We know firsthand that the Wellcome Trust is right. In October 2003, we launched our first journal, PLoS Biology, and it is thriving---not only as a destination for the best research in all areas of biology, but also as a resource for students, teachers, and members of the public who have never before had direct access to the product of scientific inquiry (see for yourself at [www.plosbiology.org](www.plosbiology.org)). We are now bringing this success and this spirit to medicine. The world of medical journals needs a fresh infusion of idealism. All of today\'s leading medical journals are more than 70 years old, and PLoS Medicine is here to challenge the status quo. We are first and foremost an open-access publisher working to ensure that everyone has access to the latest medical research and expertise. But we aim to be more than just an open-access alternative to established general medical journals. We are determined to make PLoS Medicine the best medical journal in the world by providing outstanding original research and new ideas; thought-provoking, educational, and imaginative features for readers; and the fastest, fairest, and most rigorous peer review for authors. As an open-access journal, we see our audience differently than do the conventional medical journals: our audience is composed of medical researchers, physicians, and other health-care providers, patients and their advocates, students, and the public around the world. It will be a great challenge to create a journal that will serve such a diverse audience---we welcome this challenge. We will make it possible for the results of advanced research on infectious diseases to guide treatment in remote clinics thousands of miles away. We will make the results of a clinical trial of a new drug accessible and understandable both to doctors who might prescribe it and to people who might start taking it. We will make research on rare diseases accessible to general practitioners and patients so that they can work together to recognize and treat them. Whereas some would argue that medical journals should not be accessible to patients because patients are unable to use the information effectively, we believe it is our responsibility as publishers and members of the medical community not only to give patients access, but to provide them with tools to use the medical literature wisely. Medical research is a partnership between medical scientists and millions of voluntary human participants, conducted largely with public funds. What better way to acknowledge the public\'s contribution and ensure their willingness to sponsor and participate in future research than to openly share the product of this research with them? We hope that you will enjoy reading PLoS Medicine and find it useful and provocative. Please share the journal with your colleagues, patients, and friends. Tell us what you want to see, what you like, and what we could do better. Give us your ideas for changes that will make PLoS Medicine a better journal for you and the community. Join us in reinventing the medical journal. Citation: Eisen MB, Brown PO, Varmus HE (2004) PLoS Medicine---A medical journal for the Internet age. PLoS Med 1(1): e31. [^1]: Michael B. Eisen, Patrick O. Brown, and Harold E. Varmus are the co-founders of the Public Library of Science. Michael B. Eisen is at the Lawrence Berkeley National Laboratory and the University of California, Berkeley, California, United States of America; Patrick O. Brown is at the Stanford University School of Medicine and Howard Hughes Medical Institute, Stanford, California, United States of America; Harold E. Varmus is president and chief executive of Memorial Sloan-Kettering Cancer Center, New York, New York, United States of America. E-mail: <plos@plos.org>	PubMed Central	2024-06-05T03:55:48.601777	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523847/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e31", "authors": [ { "first": "Michael B", "last": "Eisen" }, { "first": "Patrick O", "last": "Brown" }, { "first": "Harold E", "last": "Varmus" } ] }
PMC523848	DESCRIPTION of CASE {#s1} =================== A 17-year-old woman, who was born in Bangladesh, presented to an accident and emergency department in the United Kingdom with a history of being unwell for 24 hours. She had a headache and fever, and was vomiting. On questioning, there was no photophobia or neck stiffness. She lives in the United Kingdom and had been for a holiday to Bangladesh 4 months previously. The family members were all well, and none had similar symptoms. The woman had no previous medical history and was on no medication. On examination, she looked unwell, with a temperature of 38.5 °C, pulse 96 beats per minute, blood pressure 112/52 mm Hg, and a respiratory rate of 16 breaths per minute. She had a Glasgow Coma Score of 14, there was no neck stiffness, and her ocular fundi were said to be normal. There were no other significant findings on examination. Immediate investigations showed a normal blood count apart from a raised white blood cell (WBC) count of 11.1 × 10^9^ per millilitre (normal range, 4.3--10.8 × 10^9^, per millilitre), with 84% neutrophils. Her erythrocyte sedimentation rate was 15 mm in the first hour (normal range, 1--12 mm), and her C-reactive protein was less than 5 mg/l (normal range, less than 10 mg/l). A malaria screen was negative, and renal and liver function tests were normal. What Clinical Diagnoses Were Being Considered? {#s1a} ---------------------------------------------- A clinical diagnosis of probable viral meningitis was made pending the results of further investigations. Chest X ray was normal, as was an unenhanced CT brain scan. Lumbar puncture showed a normal cerebrospinal fluid (CSF) opening pressure, and the CSF was not blood stained. Laboratory analysis showed 1,606 WBC/ml (normally there are less than 5 cells/ml), of which 60% were lymphocytes with protein of 1.08 g/l (normally less than 0.6 g/l) and glucose of 2.9 mmol/l (normally greater than 50% of plasma glucose), with a corresponding plasma glucose of 5.7 mmol/l (normal range, 4--9 mmol/l). No organisms were seen on Gram stain. The CSF was negative for pneumococcal, meningococcal, and Haemophilus antigens, and bacterial culture was subsequently negative. The patient was given 2 g of ceftriaxone intravenously immediately, and ceftriaxone treatment was continued following the CSF results. Aciclovir IV, at a dose of 10 mg/kg every 8 hours, was added to cover the possibility of herpes encephalitis. What Was the Subsequent Differential Diagnosis? {#s1b} ----------------------------------------------- The differential diagnosis was now viral, bacterial, tuberculous (TB), fungal, or malignant meningitis, or sarcoidosis. Additional investigations were requested, including polymerase chain reaction (PCR) on CSF for viruses and tuberculosis. There were no risk factors for HIV infection (HIV seroconversion can present with meningitis). Mollaret's meningitis (recurrent aseptic meningitis associated with herpes simplex virus) was a possibility \[[@pmed-0010007-b1]\], though this condition characteristically presents as recurrent episodes of apparent aseptic meningitis. The following morning the patient's temperature returned to normal at 37 °C, and she felt better. Referral was made to the Infectious Disease Service. Twenty-four hours after admission, she was afebrile but clinically worse, with marked headache and vomiting. A day later she spiked a fever of 39 °C. No organisms were cultured from the CSF, urine, or blood. CSF bacterial antigens, cryptococcal antigen, and CSF auramine stain (for mycobacteria) were also all negative. The following day the patient complained of further headache, nausea, blurred vision, and photophobia. In addition, she was noted to have bilateral large pupils, which did not react to light, and very pink optic nerves. Papilloedema was thought likely, and no other neurological signs were detected. The original CSF sample was negative on PCR for tuberculosis and herpes simplex virus, C-reactive protein remained at less than 5 mg/l, and a further head CT scan with contrast was normal. The diagnosis was revised to meningoencephalitis. Other agents---such as listeria, tuberculosis (a systematic review found that PCR has a sensitivity of only 56% \[95% CI, 46%--66%\] for detecting TB meningitis \[[@pmed-0010007-b2]\]), and viruses such as others in the herpes group, mumps, and West Nile virus---were considered. The aciclovir was stopped, and quadruple tuberculosis therapy was started. What Did the Eye Signs Mean? {#s1c} ---------------------------- The fixed, dilated pupils were of major concern, and urgent ophthalmic review was requested. Examination by the ophthalmologist showed reduced vision at 6/60 right eye and 6/36 left eye. On testing with Ishihara charts, the patient had severely reduced colour vision. Her pupils were large and non-reactive to light or accommodation in both eyes. The eyes were inflamed, with cells present in the aqueous and vitreous humours. The optic nerves were swollen and very pink ([Figure 1](#pmed-0010007-g001){ref-type="fig"}), but there was normal spontaneous venous pulsation present---demonstrating that this was not papilloedema (see [Video 1](#pmed-0010007-v001){ref-type="other"}). Bilateral choroidal infiltrates with overlying serous retinal detachments were also present. ::: {#pmed-0010007-g001 .fig} Figure 1 ::: {.caption} ###### Fundal Appearance of the Patient\'s Eye The large arrow indicates the pink optic nerves; the star shows localised retinal detachment; and the small arrow pointing down shows small, white choroidal granulomas. ::: ![](pmed.0010007.g001) ::: ::: {#pmed-0010007-v001 .fig} Video 1 ::: {.caption} ###### Spontaneous Venous Pulsation of the Veins at the Optic Nerve Head ::: ::: What Was the Final Diagnosis and Treatment? {#s1d} ------------------------------------------- The combination of the clinical symptoms, signs, and ocular features was characteristic of Vogt-Koyanagi-Harada (VKH) syndrome \[[@pmed-0010007-b3]\]. All antibiotics were stopped, and high-dose corticosteroids were started at 100 mg prednisolone daily. At one week there was no significant ocular improvement, although the patient\'s headache and vomiting were now gone, and she felt much better. Additional immunosuppressive therapy was initiated with cyclosporin and mycophenolate, and within a further week the patient\'s vision started to improve, with settling of the ocular signs. Oral corticosteroids were tapered, as was the cyclosporin, and by one month the patient\'s vision had returned to normal, and the ocular signs continued to settle. By three months the cyclosporin was discontinued, the steroid dose was reduced to 5 mg daily, and the mycophenolate dose was tapered. By six months all therapy was discontinued. At review six months later, the patient remained well, with normal vision and normal optic nerves. DISCUSSION {#s2} ========== This young patient presented acutely with a fever and some signs suggestive of meningitis. She was initially treated as having viral meningitis, but the CSF findings indicated that other aetiologies needed to be considered. In particular, in a woman who previously lived in and recently visited Bangladesh, with a lymphocytic meningitis and borderline CSF glucose, tuberculosis had to be considered. Initially the ocular symptoms and signs were not a prominent feature, but the signs were likely to have been present when the patient was first seen. The typical CSF changes associated with meningitis of different aetiologies are shown in [Table 1](#pmed-0010007-t001){ref-type="table"}. In this case the mixed lymphocytes and neutrophil leucocytosis with a borderline CSF glucose on the patient\'s initial CSF sample were consistent with bacterial or TB meningitis. Viral infection was far less likely, as only mumps is consistently associated with reduced CSF glucose. ::: {#pmed-0010007-t001 .table-wrap} Table 1 ::: {.caption} ###### CSF Changes in the Most Commonly Encountered Types of Meningitis ::: ![](pmed.0010007.t001) ::: Infectious Causes of Meningitis {#s2a} ------------------------------- There is a wide range of infectious causes of meningitis worldwide. The likely infecting organism will be determined by the age and immune status of the patient plus the situation in which the infection was contracted. Thus, in an immunocompetent adult in the UK, enteroviruses are the commonest cause of viral meningitis, with meningococcus and pneumococcus the commonest bacterial agents. Tuberculosis is more common in people who have lived in a highly endemic area. In other parts of the world, the differential diagnosis may include viral infections such as West Nile virus in the continental United States and Japanese B encephalitis in Asia, or other pathogens such as rickettsiae, borrelia (Lyme disease), and protozoa. In the immunocompromised host, listeria must be considered, and there is an increased risk of fungal infection and tuberculosis. Finally, it is important to consider sexual exposure, as both secondary syphilis and HIV seroconversion may present with meningitis. It is important, therefore, in the evaluation and management of patients presenting with a meningoencephalitis that the differential diagnosis be continually reviewed if the patient is not responding to therapy ([Table 2](#pmed-0010007-t002){ref-type="table"}). When appropriate investigations have been performed and are negative and symptoms persist, non-infective causes of CSF inflammation must be considered---as turned out to be the case here ([Table 3](#pmed-0010007-t003){ref-type="table"}). ::: {#pmed-0010007-t002 .table-wrap} Table 2 ::: {.caption} ###### What to Do When the Patient Is Not Getting Better ::: ![](pmed.0010007.t002) ::: ::: {#pmed-0010007-t003 .table-wrap} Table 3 ::: {.caption} ###### Non-Infectious Causes of Abnormal CSF ::: ![](pmed.0010007.t003) ::: VKH Syndrome {#s2b} ------------ VKH syndrome \[[@pmed-0010007-b4],[@pmed-0010007-b5],[@pmed-0010007-b6]\] is a systemic disease involving various melanocyte-containing organs. It is rare in white Northern Europeans and white Americans but much more common in people with darker, pigmented skin. For example, among patients presenting with uveitis, about one in ten in Japan and one in 50 in India have VKH syndrome \[[@pmed-0010007-b6],[@pmed-0010007-b7]\]. It presents acutely with varying symptoms and signs, which include meningoencephalitis, visual blurring, and deafness. The eye signs are very characteristic and can help to make the diagnosis. The most prominent ocular finding is intensely pink optic nerves (see [Figure 1](#pmed-0010007-g001){ref-type="fig"}), with severe visual reduction and loss of function, which accounts for the absent pupillary responses. VKH syndrome is usually bilateral, but occasionally the eyes can be affected asymmetrically so that one is very mildly involved. The syndrome is accompanied by marked intraocular inflammation, and there are choroidal infiltrates ([Figure 2](#pmed-0010007-g002){ref-type="fig"}) associated with serous retinal detachments, which may be localised ([Figure 3](#pmed-0010007-g003){ref-type="fig"}) or affect the whole retina ([Figure 4](#pmed-0010007-g004){ref-type="fig"}). It is likely that these detachments are due to the retinal pigment epithelium (RPE) being affected by the underlying inflammatory choroidal granulomas (which heal leaving scars; see [Figure 5](#pmed-0010007-g005){ref-type="fig"}), and fluid accumulates underneath the retina because of reduced function of the RPE when it becomes inflamed. ::: {#pmed-0010007-g002 .fig} Figure 2 ::: {.caption} ###### White Choroidal Infiltrates (Arrow) Seen in VKH Syndrome with Very Pink Optic Nerve Head ::: ![](pmed.0010007.g002) ::: ::: {#pmed-0010007-g003 .fig} Figure 3 ::: {.caption} ###### Localised Retinal Detachment ::: ![](pmed.0010007.g003) ::: ::: {#pmed-0010007-g004 .fig} Figure 4 ::: {.caption} ###### Total Retinal Detachment, Where Whole Retina is Grey in Colour ::: ![](pmed.0010007.g004) ::: ::: {#pmed-0010007-g005 .fig} Figure 5 ::: {.caption} ###### Scarring When Choroidal Granulomas Subside ::: ![](pmed.0010007.g005) ::: The disorder is caused by an immune response to melanin and affects parts of the body where melanin is found. The initiating stimulus for this response is unknown, but T-cells sensitised to melanin-associated antigens are found in the peripheral blood. In the ear, the melanocytes of the inner ear are the target, and the inflammatory response here results in hearing loss and balance problems. In longstanding untreated cases, depigmentation may occur in other sites such as skin (vitiligo; [Figure 6](#pmed-0010007-g006){ref-type="fig"}) and eyelashes (poliosis; [Figure 7](#pmed-0010007-g007){ref-type="fig"}), but these are uncommon when corticosteroids and other immunosuppressive agents are used in treatment. Depigmentation of the RPE can also occur, giving a 'sunset' appearance to the dark fundus. ::: {#pmed-0010007-g006 .fig} Figure 6 ::: {.caption} ###### Vitiligo on Skin of Forearm ::: ![](pmed.0010007.g006) ::: ::: {#pmed-0010007-g007 .fig} Figure 7 ::: {.caption} ###### Poliosis Note white eyelashes on child. ::: ![](pmed.0010007.g007) ::: Treatment with high-dose corticosteroids is essential \[[@pmed-0010007-b3]\] and should be initially 1--2 mg/kg/day. This treatment can be given orally or intravenously, depending on how unwell the patient is. However, patients commonly need other immunosuppressive agents as well, so as to allow the dose of steroids to be reduced more quickly. Both cyclosporin and mycophenolate are very useful as steroid-sparing agents, with cyclosporin having the advantage of a variable-dose regimen for more rapid onset of action. On the down side, cyclosporin can cause hirsutism, especially in combination with corticosteroids (which can also cause this side effect). Unfortunately, the costs of cyclosporin and mycophenolate may preclude their use in resource-poor settings, with the result that patients may require high-dose corticosteroids for much longer, with all the concomitant side effects. Inadequate initial treatment may increase the risk of recurrence and long-term complications. Treatment is required until the disease goes into remission. The meningoence-phalitic signs and retinal and choroidal signs settle quickly, often within a week or so, whereas the optic nerve inflammation may take longer to settle. The visual prognosis depends on the degree of permanent damage to the optic nerve and the macula area, which often shows considerable pigment clumping as a result of the damage to the RPE. Relapse affecting the optic nerve, choroids, and retina is uncommon, provided that treatment has been given for long enough. However, recurrent anterior uveitis requiring steroid drops is common. This is not a threat to sight if adequately controlled. As with any other cause of intraocular inflammation particularly associated with choroidal involvement, VKH syndrome can lead to reduced vision via cataracts, glaucoma damaging the optic nerve, and new vessels growing into the retina through the damaged RPE (choroidal neovascular membrane). ### Key Learning Points Consider meningitis in the differential diagnosis of a patient presenting with fever and headache.CSF analysis is essential to confirm meningitis and as part of establishing the cause.Consider non-infectious causes when a patient does not respond rapidly to therapy.Blurring of vision must be investigated and may help in determining the underlying diagnosis or the presence of papilloedema. ### Suggested Reading Citation: Lynn W, Lightman S (2004) Fever, headache, and visual blurring in a 17-year-old woman. PLoS Med 1(1): e7. CSF : cerebrospinal fluid PCR : polymerase RPE : retinal pigment epithelium TB : tuberculous VKH : Vogt-Koyanagi-Harada WBC : white blood cell [^1]: William Lynn is a consultant in infectious diseases and the medical director at Ealing Hospital, London, United Kingdom. E-mail: <William.lynn@eht.nhs.uk>. Susan Lightman is professor of Clinical Opthalmology and is head of the Department of Ophthalmology at Moorfields Eye Hospital, London, United Kingdom. E-mail: <s.lightman@ucl.ac.uk> [^2]: Competing Interests: The authors are on the editorial board of PLoS Medicine.	PubMed Central	2024-06-05T03:55:48.603176	2004-10-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523848/", "journal": "PLoS Med. 2004 Oct 19; 1(1):e7", "authors": [ { "first": "William", "last": "Lynn" }, { "first": "Sue", "last": "Lightman" } ] }
PMC523849	Background ========== MicroRNAs (miRNAs) are small, \~18--24 nt. noncoding RNAs that are found in all eukaryotes and are cleaved from larger \~70 nt. precursors via the action of Dicer enzyme \[reviews: ref. \[[@B1],[@B2]\]\]. MicroRNAs are thought to degrade messenger RNAs via eliciting mRNA degradation (if they bind in a perfect or near-perfect complementarity to the target mRNA), or to arrest translation of the mRNAs (if the binding complementarity is imperfect). Although a number of microRNA targets have been identified in plants, C. elegansand Drosophila\[[@B1],[@B2]\], only one mammalian microRNA target has yet been validated \[[@B3],[@B4]\]. Five different papers have recently appeared that used computational approaches to predict microRNA targets in Drosophila\[[@B5]-[@B7]\], and mammals \[[@B8],[@B9]\]. These studies only considered hits occurring within 3\'-UTR regions that were conserved across related species, and favored or required a short region of perfect complementarity towards the 5\'-end of microRNAs. However, there is reason to suspect that the rules governing microRNA-target interactions may not be universal. For example, in plants, most of the known microRNAs bind in a perfect or near-perfect manner to mRNA targets located within the protein coding region (cds) \[[@B10],[@B11]\]. In contrast, in C. elegans\[[@B12]\] and Drosophila\[[@B13]\], known microRNAs lack long stretches (\>10) of complementarity with their targets and generally interact within the 3\'-untranslated region (3\'-UTR). Furthermore, whereas the 5\'-ends of many DrosophilamicroRNAs recognize 5--6 nt. common motifs within the target, these motifs are not a general feature of mammalian microRNAs \[[@B14]\]. Thus, it is conceivable that human microRNA targets do not follow the same constraints as observed in C. elegansand Drosophila. In the present paper, we have performed an unbiased statistical analysis of the manner in which human microRNAs interact complementarily with human mRNAs present in the NCBI human RefSeq database, looking for characteristics that differ significantly as compared with scrambled versions of the same microRNA sequences. The results demonstrate several novel features of human microRNA-mRNA interactions that differ from C. elegansand Drosophila, and identify a short-list of promising candidate microRNA-mRNA target pairs that are unlikely to have arisen by chance. Results ======= Population-wide statistical analyses were first carried out by examining the types of complementary interactions that occur between the set of microRNAs listed in Lagos-Quintana et al \[[@B15]\], and the set of human RefSeq mRNAs downloaded in August 2003. To obtain a fuller list of outlier mRNAs, analyses were repeated using all human microRNAs listed on the Sanger microRNA repository \[[@B16]\] and the set of human RefSeq mRNAs listed as of December 2003 \[[@B17]\]. To define the types of interactions that can occur by chance, ten independent sets (\"replications\") of scrambled microRNA counterpart sequences were generated and examined for complementarity with the mRNA population. Our underlying assumption is that scrambled sequences will hit mRNA at random and define the \"noise\" level in any given situation, whereas microRNA sequences will hit the same number of \"noise\" interactions plus any true targets. Unless otherwise noted, the scrambled sequences were random permutations of the microRNA sequences, keeping constant the overall nucleotide composition. Because microRNAs have a distinctive nonrandom di-nucleotide composition, we also confirmed that key findings were obtained when using scrambled sequences that had similar di-nucleotide composition to the microRNAs. 1. Human microRNAs tend to have longer exact hits upon mRNAs than do their scrambled counterparts ------------------------------------------------------------------------------------------------- First, we characterized the length distribution of exact complementarity between the population of mRNAs vs. the set of nonredundant microRNAs (i.e. those that overlapped by 10 or more bases were collected into groups and the longest member of the group was chosen as nonredundant). MicroRNAs produced significantly longer exact \"hits\" on mRNAs than their scrambled counterparts when G:U matches were excluded (fig. [1](#F1){ref-type="fig"}). There was an excess number of hits in the microRNA set relative to scrambled control sequences at all exact hit lengths (10 or greater), and the difference became proportionately greater at longer hit lengths (see below). When microRNAs were compared to scrambled sequences that matched the di-nucleotide composition of microRNAs, similar results were obtained. In contrast, this trend was not observed when G:U matches were included (not shown). Experimental studies suggest that RNA interference and arrested translation can still be elicited when small RNAs are modified to replace a number of Watson-Crick base pairs by G:U matches \[[@B18]-[@B20]\]. On the other hand, G:U matches have distinctive binding energy and spatial orientation \[[@B21]\]. Unless otherwise qualified, \"exact hits\" will refer to complementarity without G:U matches. 2. Constructing an outlier set of microRNAs based on cut-offs of exact hit length, gapped BLAST score and presence of multiple hits ----------------------------------------------------------------------------------------------------------------------------------- As shown in figure [1](#F1){ref-type="fig"}, there are a total of 101, 279 microRNA hits upon RefSeq sequences hitting exactly ≥10 bases in a row, compared to 75,031 hits produced by scrambled microRNA sequences. The difference (26, 248 hits, distributed among 8258 mRNA sequences) is highly significant (p = 3 × 10^-9^) and suggests that about 1/4 of the total hits in this \"10+ set\" occur upon \"true\" biological mRNA targets. Our approach is to identify further the mRNAs that represent statistical outliers (i.e. that are unlikely to be hit by chance) within this larger \"10+ set\" by comparing properties of hits made by the set of microRNA sequences vs. the set of scrambled sequences. At any given parametric value, the number of hits observed in the microRNA set, minus the number of hits in the scrambled set, provides an estimate of the number of true microRNA targets that satisfy that parametric value. We examined three different hit properties -- a) exact hit length, b) gapped BLAST score and c) presence of multiple hits -- both alone and combined with each other. Starting from the \"10+ set\" estimated to contain only 26% true targets (see above), we added additional criteria to compile a list of candidates estimated to contain over 80% true targets. ### a) Exact hit length The most important single parameter for discriminating hits produced by microRNAs vs. scrambled sequences appears to be exact hit length. At a cut-off of 17 exact hit length, there were 14 mRNAs hit by the microRNA set that satisfied this criterion, vs. an average of 1.9 mRNAs hit by each of the scrambled sequence sets (fig. [1](#F1){ref-type="fig"}). Stated another way, this criterion gives a discrimination ratio of 7.4 to 1. A similar discrimination ratio was observed when comparing scrambled sequences maintaining the same di-nucleotide composition as the microRNAs. ### b) Gapped-BLAST score Overall complementarity of the microRNA-mRNA alignment was also examined within the \"10+ set\" of individual mRNAs exhibiting exact microRNA hits of at least 10 bases in a row. A modified gapped-BLAST algorithm \[[@B22]\] was used to compute the optimal alignment, employing a weighted score that takes gaps and mismatches into account (r = 10, q = -2.5, G = 8, E = 0.5). Although the two curves overlap quite a bit, their means are significantly different from each other (p \< 0.0001), and the microRNA distribution exhibits a discrete \"tail\" at higher scores that differs significantly from the scrambled distribution (fig. [2](#F2){ref-type="fig"}). ### c) Multiple hits In lower organisms, individual validated microRNA targets tend to receive multiple hits by distinct microRNAs \[[@B1],[@B2]\]. mRNA sequences within the \"10+ set\" were hit by multiple nonredundant microRNAs more often than by their scrambled counterparts, and this was particularly striking when the hits were located close together (fig. [3](#F3){ref-type="fig"}). ### d) Combining parameters When combined, all three parameters (exact hit length, gapped-BLAST scores and multiple hits) gave better discrimination power than using any single feature, supporting the idea that they are relevant to identifying biologically relevant mRNA targets. We examined three different combinations of parameter cut-off values: 1) One combination consisted of targets with multiple hits from distinct microRNAs less than 25 bases apart, with at least one exact hit ≥13 bases and with at least one gapped BLAST score ≥185 (not counting G:U). For the next two lists, we scored only exact hits ≥10 bases long and that occurred ≤50 times within the entire mRNA population; this minimized \"noise\" arising from common or low-complexity target sequences, albeit at the cost of removing some target sequences that are shared within protein families. 2) Criteria required two or more hits from distinct microRNAs ≤100 bases apart, at least one exact hit ≥14 bases and one gapped-BLAST score of ≥190 (not counting G:U). 3) This required hits ≤500 bases apart, at least one exact hit ≥14 bases, and at least one gapped-BLAST score \> 89% of the best-possible score including G:U matches (this takes into account the fact that longer microRNAs have greater possible absolute scores than shorter microRNAs). All three approaches produced lists of outlier mRNAs that had overlapping members, shared similar characteristics and exhibited similar discrimination ratios. For simplicity and robustness, these have been combined (together with the candidates identified by exact hit length alone) into a single list consisting of 71 outlier mRNAs (Table [1](#T1){ref-type="table"}). The combined list was hit by almost the entire set of nonredundant microRNAs (i.e., 107 out of 109). In contrast, scrambled counterpart sequences hit an average of 13.7 ± 1.15 targets and were represented by 54.3 ± 3.5 nonredundant sequences. The combined outlier set gives an overall discrimination ratio of 5.2 to 1, meaning that 57 of the 71 mRNAs are in excess of the number that could be reasonably expected by chance, hence should be accurately assigned as true targets for one or more microRNAs. See <http://genomebiology.com/2004/5/2/P4> for additional data files including a fully annotated outlier mRNA set, a list of all microRNA hits upon this set (extended with and without including G:U matches), and a list of the nonredundant microRNAs together with their putative mRNA targets. 4. Characterizing the mRNA outlier set -------------------------------------- The 71 mRNAs in the outlier set had a larger number of microRNA hits per kilobase of target sequence than did the scrambled sequences (2.18 ± 0.1 vs. 1.83 ± 0.085, p = 0.006). As well, individual microRNAs hit multiple (up to 17) distinct members of the outlier set, which again happened significantly more often than by chance (fig. [4](#F4){ref-type="fig"}). These findings indicate that the outlier mRNAs are different as a whole from the mRNAs that were hit by scrambled counterparts, even those that satisfied the same cut-off criteria. The outlier mRNA set contained very similar types of targets as predicted by previous computational studies \[[@B5]-[@B8]\], including members of the same gene families. For example, Lewis et al. \[[@B8]\] described E2F1 as a candidate target whereas we found E2F6 (Table [1](#T1){ref-type="table"}). Transcription factors (including homeobox genes) and nucleic acid-binding proteins are among the top predicted microRNA targets. As well, many other functional categories are represented including kinases, receptors and other signal transduction proteins, membrane and cytokeletal proteins, and effectors of differentiation (Table [1](#T1){ref-type="table"}). However, surprisingly, we found that the human candidate microRNA target list also had some features that differed significantly from the known targets in C. elegansand Drosophila. For example, there was [no]{.underline} preference for microRNA hits to be located within 3\'-untranslated regions: 5% of hits were located in the 5\'-UTR, 1% at the 5\'-UTR/coding junction, 67% in the protein coding region, 1% at the coding/3\'-UTR junction, and only 26% in the 3\'-UTR. This distribution was not significantly different from hits produced by the scrambled sequences. As well, the best microRNA hits upon candidate mRNA targets did [not]{.underline} have relatively better target complementarity near their 5\'-end: Only 13% of hits had ≥ 7 exact hit length starting at position 1 or 2 relative to the 5\'end of the microRNA (vs. 17.5% of hits produced by scrambled sequences). MicroRNA 145 is particularly interesting because it hits 17 distinct targets on the candidate list, of which a disproportionate number (6) are in the signal transduction category and three of these are related to GTPase activation (Rho GTPase-activating protein (RICS), G protein gamma 7, and hypothetical protein FLJ32810 -- containing RhoGAP and SH3 domains; Table [1](#T1){ref-type="table"}). A recent study showing that miR-143 and miR-145 are both underexpressed in colorectal neoplasia \[[@B23]\] had previously proposed the first two of these candidates as potential targets. Interestingly, the third target found here is not only novel (XM\_350859, RhoGAP-like) but is hit by both miR-143 and miR-145 in close proximity (see additional data file 2 in <http://genomebiology.com/2004/5/2/P4>), further suggesting that this is likely to be a true biological target for microRNA regulation. Discussion ========== By comparing how the population of microRNAs vs. their scrambled counterparts interact with the population of human RefSeq mRNA sequences, we estimate that the probability of detecting a true microRNA target increases a) as the length of exact complementarity of a \"hit\" between microRNA and target increases, b) as the overall complementarity of a \"hit\" increases (allowing for gaps, mismatches and G:U matches), and c) as two or more distinct microRNAs hit the same mRNA in closer proximity. Targets in the outlier mRNA set also received more hits per unit length and more multiple hits from distinct microRNAs than expected by chance. Finally, we found cases in which an individual microRNA hit multiple mRNAs that belonged to the same functional class. The analysis suggests that target complementarity is a major factor in identifying biologically relevant mRNA targets: As values of each parameter increase, the difference between the number of hits in the microRNA set vs. the scrambled set increases steadily, and combining all three parameters gives better discrimination power than using any single feature. So far, these conclusions agree with five different papers that used computational approaches to predict microRNA targets in Drosophila\[[@B5]-[@B7]\], and mammals \[[@B8],[@B9]\], using different strategies, criteria and filters than employed here. However, three significant differences were observed between human mRNAs in the outlier set and Drosophilatargets: 1) Human microRNAs hit mRNAs with exact hit lengths extending much longer than observed in Drosophila, up to and including perfect complementarity. 2) Human microRNAs hit candidate mRNA targets within the protein coding region about 2/3 of the time. (This resembles the manner in which plant microRNAs hit their mRNA targets \[[@B10],[@B11]\].) 3) The stretches of perfect complementarity within microRNA hits in the outlier mRNA set were not biased to occur near the 5\'-end of the microRNA. This is not necessarily at odds with earlier analyses, since our outlier set includes only perfect stretches of 13 bases or more, and the 5\' end may be more critical in those cases where only a short perfect stretch of complementarity exists. One might object that our ability to detect certain trends seen in Drosophilaand C. eleganswas simply obscured by the fact that we searched the large sequence space represented by all human mRNA sequences -- the larger the sequence space, the greater the chance that any given target criterion will be satisfied by scrambled sequences, hence the more difficult it can be to detect true targets above the noise level. We agree that this can be a problem using very large sequence databases, such as the human EST database or the entire human genome. As well, using cut-off levels of parameter distributions to define the candidate list probably excludes many true human mRNA targets. However, human RefSeq was demonstrably not too large for our analysis, since very strong trends were observed in a variety of other parameters (figs. [1](#F1){ref-type="fig"},[2](#F2){ref-type="fig"},[3](#F3){ref-type="fig"},[4](#F4){ref-type="fig"}). Based upon sequence complementarity, at least 57 out of the 71 members of the outlier set are predicted to represent true microRNA targets (Table [1](#T1){ref-type="table"}). Indeed, since this paper was first submitted for publication, one of the mRNAs on this list, HOXB8, has been experimentally confirmed \[[@B24]\]. Note, however, that accessory factors in the RISC might also help to determine which potential mRNA targets will actually be sites of regulation in vivo. As well, microRNA and target must be expressed in the same times and places in adequate concentrations; secondary structure of the mRNA target region may be important \[[@B19],[@B20]\]; see also \[[@B8]\]; and RNA A-to-I editing \[[@B25],[@B26]\] might operate to prevent certain target sequences from binding microRNAs adequately. Conclusions =========== In summary, the population-wide characteristics of microRNA-mRNA sequence complementarity indicate that microRNAs recognize a subset of human mRNA sequences better than expected by chance. This outlier set does obey a number of properties expected for true biological mRNA targets, but does not show a bias for target regions to be located within the 3\'-UTR of the mRNA, and stretches of perfect complementarity are not biased towards the 5\'-end of the microRNA. If the candidate list is representative of the full set of biologically significant targets, then the total number of mRNA targets in humans may be much greater than previously proposed \[[@B8]\]. Abbreviations ============= 5\'-UTR, 5\'-untranslated region. CDS, protein coding region. 3\'-UTR, 3\'-untranslated region. Methods ======= MicroRNAs --------- Statistical analyses were first carried out using the set of mouse and human microRNAs listed in Lagos-Quintana et al \[[@B15]\], and then repeated to obtain individual candidate mRNA targets using all human microRNAs listed on the Sanger microRNA repository \[[@B16]\] as of December 2003. These sources were combined to create nonredundant microRNA sets (i.e. microRNAs that have 10 or more consecutive nucleotides in common were collected into groups and the longest member of the group was chosen as nonredundant). Almost all mouse microRNAs have exact human counterparts, but hits were annotated with mouse entries in cases of minor corrections and discrepancies between these two sources. One individual microRNA (mir-207) and several scrambled sequences were found to be low-complexity or complementary to abundant repeats (e.g., Alu) and were removed from consideration. mRNAs ----- Analyses were first carried out using the set of human RefSeq mRNAs available in August 2003, and then supplemented with additional human RefSeq mRNAs listed as of December 2003. A) Sequences in RefSeq \> 20,000 bases long were removed from consideration because they were hit by many, if not all microRNAs, and a few sequences \> 15,000 bases long were removed from the final candidate list because they had a relatively high false-positive probability. B) When counting the number of hits over the population of mRNAs, two hits were counted as redundant if the entire region around the hit (plus or minus 25 nucleotides on each side) was identical. C) When counting distinct hits by microRNAs on the same target, two hits were counted as redundant if they shared the same exact hit. This minimized possible artifacts due to overlapping microRNAs, as well as removed cases in which microRNAs hit exactly-repeating sequences within the target. D) In tabulating hits onto mRNA targets, we did not count hits that contained low-complexity sequences as detected by the DUST algorithm encoded by a Perl script provided by Lincoln Stein \[[@B27]\]. E) When assembling the candidate mRNA target list, we chose a single exemplary mRNA and removed other entries that were transcript variants or nearly identical by BLAST searching. In the course of this study, some of the target mRNAs were removed from RefSeq for routine genome annotation processing. If these were subsequently replaced with updated versions of these mRNAs in RefSeq that included the same hits, the latter version is listed here as well. For those entries removed but not replaced in RefSeq at the time of submission of the manuscript, other active entries currently in Genbank are listed if possible. Statistics ---------- To decide whether the number of observed microRNA hits were significantly different from chance, 10 replications of scrambled sequences were used to estimate prediction intervals. The prediction interval allows one to say with 95% confidence that any single new replication of the scrambled set will be below the value of the microRNA set. Prediction intervals were chosen as more conservative and more appropriate than confidence intervals. Authors\' contributions ======================= NS contributed biological expertise, whereas VT contributed statistical and computational expertise. The analyses were carried out together, and both authors read and approved the final manuscript. Acknowledgements ================ Supported by NIH grants DA15450 and LM07292. This Human Brain Project/Neuroinformatics research is funded jointly by the National Library of Medicine and the National Institute of Mental Health. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### microRNAs and their scrambled counterparts interact differently with the population of human mRNAs.Shown are all exact hits ≥ 10 bases long (not counting G:U matches) produced on human RefSeq mRNAs by the set of nonredundant microRNAs, vs. the average of 10 replications of scrambled control sequences. Shown is the number of hits as a function of exact hit length. Only the longest hit was counted: e.g., for a hit of length 18, the two subsets of length 17 in the same hit position were not counted. ::: ![](1471-2105-5-139-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Distribution of gapped-BLAST scores in hits made by microRNAs and scrambled counterparts.Without permitting G:U matches in the extension phase, the microRNAs had better average gapped-BLAST scores than scrambled counterparts across all mRNAs in the \"10+ set\" (153.00 ± 0.03 vs. 150.98 ± 0.01, mean ± s.e.m., p \< 0.0001). With permitting G:U matches in the extension phase, the microRNA set showed significantly fewer G:U matches overall relative to scrambled counterparts, even when holding constant the length of the exact hit (2.891 ± 0.004 vs. 2.939 ± 0.001, p \< 0.0001). ::: ![](1471-2105-5-139-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Number of distinct mRNA sequences which received hits from two or more distinct microRNAs, as a function of the minimum distance between hits.Distance of 0 or 1 was excluded because this might be produced by partial overlap of microRNA sequences. ::: ![](1471-2105-5-139-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Individual microRNAs hit multiple targets on the candidate list, more often than expected by chance. ::: ![](1471-2105-5-139-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The Outlier mRNA Set (Candidate Target List) ::: [Transcription factors and other nucleic-acid binding proteins (15)]{.underline} ------------------------------------------------------------------------------------------------------------------------------------------------------ homeo box B8 (HOXB8) E2F transcription factor 6 (E2F6) transcription factor 20 (AR1) (TCF20) DEAD (Asp-Glu-Ala-Asp) box polypeptide 51 (DDX51) similar to ATP-dependent RNA helicase DDX24 (DEAD-box protein 24) (LOC221311) myeloid/lymphoid or mixed-lineage leukemia (trithorax homolog); translocated to, 1 (MLLT1) high mobility group AT-hook 2 (HMGA2) polymerase (DNA directed), theta (POLQ) strand-exchange protein 1 (SEP1) hypothetical protein FLJ12994 -- RFX DNA-binding domain similar to LINE-1 reverse transcriptase homolog (LOC285907) similar to hypothetical protein (L1H 3 region) -- related to reverse transcriptase similar to putative p150 (LOC282945) -- related to reverse transcriptase similar to reverse transcriptase related protein (LOC222252) similar to RTl1 (LOC376283) -- related to reverse transcriptase [Kinases, receptors and other signaling proteins (13)]{.underline} fyn-related kinase (FRK) WNK kinase, lysine deficient 3 (PRKWNK3) protein phosphatase 2, regulatory subunit B (B56), epsilon isoform (PPP2R5E) EphA5 receptor (EPHA5) killer cell lectin-like receptor subfamily A, member 1 (KLRA1) polycystin and REJ (sperm receptor for egg jelly homolog, sea urchin)-like (PKDREJ) integrin, alpha X (antigen CD11C (p150), alpha polypeptide) (ITGAX) inositol 1,4,5-triphosphate receptor, type 1 (ITPR1) hypothetical protein FLJ32810 -- RhoGAP domain, SH3 domain hypothetical protein FLJ00058 -- G protein gamma 7 Rho GTPase-activating protein (RICS) hypothetical protein FLJ30899 -- probable ras GAP similar to ADP-ribosylation factor-like membrane-associated protein (LOC132946) -- ARF-like small GTPase domain, Sar1p-like member of the Ras-family [Membrane and extracellular proteins (11)]{.underline} Laminin, beta 4 (LAMB4) laminin, gamma 2 (LAMC2) fibronectin 1 (FN1) collagen, type IV, alpha 5 (Alport syndrome) (COL4A5) collagen, type XIX, alpha 1 (COL19A1) similar to Voltage-dependent anion-selective channel protein 1 (VDAC-1) ATPase, Na+/K+ transporting, alpha 2 (+) polypeptide (ATP1A2) complement component 1, q subcomponent, beta polypeptide (C1QB) hypothetical protein FLJ20506 -- transmembrane protein MAM domain containing glycosylphosphatidylinositol anchor 1 (MDGA1) -- Ig, MAM domains similar to TCAM-1 (LOC284171) [Cytoskeletal domain-containing proteins (7)]{.underline} myosin heavy chain Myr 8 (MYR8) ankyrin repeat domain 17 (ANKRD17) KIAA1817 protein -- intermediate filament, ATPase, PDZ, Band 4.1, FERM domains chromosome 10 open reading frame 39 (C10orf39) -- homologous to myosin, plectin oxysterol binding protein 2 (OSBP2) -- pleckstrin homology domain KIAA1202 protein -- PDZ, ATPase domains hypothetical protein FLJ23529 -- homolgous to dynein heavy chain [Miscellaneous or unknown function (26)]{.underline} cell cycle progression 2 protein (CPR2) olfactomedin 3 (OLFM3) histidine rich calcium binding protein (HRC) interferon-related developmental regulator 1 (IFRD1) KIAA1301 protein -- NEDD4-related E3 ubiquitin ligase NEDL2 KIAA1203 protein -- ubiquitin C-terminal hydrolase hydroxyprostaglandin dehydrogenase 15-(NAD) (HPGD) UDP-GlcNAc:betaGal beta-1,3-N-acetylglucosaminyltransferase 1 (B3GNT1) KIAA1854 protein -- leucine rich repeat C-terminal domains testis specific, 14 (TSGA14) chromosome 4 open reading frame 1 (C4orf1)-membrane AND nuclear protein hypothetical protein FLJ33069 hypothetical protein FLJ38464 hypothetical protein LOC285431 hypothetical protein LOC284107 similar to agCP1362 \[Anopheles gambiae str. PEST\] (LOC344751) KIAA1632 protein similar to hypothetical protein D11Ertd497e (LOC343360) LOC138724 LOC343460 LOC340963 LOC343220 LOC285842 LOC352767 LOC350293 :::	PubMed Central	2024-06-05T03:55:48.604771	2004-9-28	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523849/", "journal": "BMC Bioinformatics. 2004 Sep 28; 5:139", "authors": [ { "first": "Neil R", "last": "Smalheiser" }, { "first": "Vetle I", "last": "Torvik" } ] }
PMC523850	Background ========== Since the publication of draft sequences for the human \[[@B1]\] and mouse \[[@B2]\] genomes, several groups have run large-scale comparisons of the sequences to detect regions of conserved sequence. An initial survey of these was published along with the draft mouse genome \[[@B2]\], with additional comparisons appearing since then \[[@B3]\]. Briefly, protein coding genes are -- as we might expect -- among the most strongly conserved regions, but homologous sequences can be found throughout the genome. In total, it is possible to align up to 40% of the mouse genome to human sequence \[[@B4]\], but it seems likely that at least some of this is just random \"comparative noise\" -- regions of sequence which serve no particular purpose but which, purely by chance, have not yet accumulated enough mutations to make their evolutionary relationship unrecognisable. However, it is widely accepted that some of the noncoding-but-similar regions, especially those with the highest levels of sequence identity between the two species, are preferentially conserved because they perform some important function. It has been estimated that around 5% of the genome is under purifying selection \[[@B2]\], indicating that mutations in these regions have deleterious effects: a strong suggestion of some important function. Here, we apply the Eponine Windowed Sequence (EWS) sequence analysis method method which uses a Relevance Vector Machine (RVM) \[[@B5]\] to extract a minimal set of short motifs which are able to discriminate between two sets of sequences: in this case, a positive set of conserved non-coding sequences and a negative set of randomly picked non-coding sequences. The EWS model is an adaption of the Eponine Anchored Sequence (EAS) model, first applied for transcription start site prediction in \[[@B6]\] and subsequently used to predict a range of additional biological features including translation start sites and transcription termination sites \[A. Ramadass, unpublished\] While EAS is designed to classify individual points in a sequence -- a feature which allows the model to predict precise locations for features such as transcription start sites -- EWS classifies complete blocks (windows) of sequence. The basis functions (inputs) of the RVM are sums of position-weight matrix scores \[[@B7]\] across the whole window. Results ======= We considered a set of alignments made by the blastz program \[[@B4]\] between release NCBI33 of the human genome and release NCBIM30 of the mouse genome. Since unprocessed blastz aligns around 40% of human sequence to the mouse genome, we chose to focus on the \'tight\' alignments. These are a subset of alignments which are rescored and thresholded using a set of parameters given in \[[@B4]\], and cover only around 5.6% of the human genome -- a proportion much closer to the fraction of bases thought to be under purifying selection \[[@B2]\]. In total, the tight blastz set contained 787173 blocks of sequence with high-scoring alignments between the two genomes. We considered only those blocks assigned to human chromosome 6, a 170 Mb chromosome which has recently undergone manual annotation of gene structures and other features \[[@B8]\]. This chromosome included 44105 (5.6%) of the total alignments. These varied in length from 34 to 9382 bases, with a length distribution skewed towards relatively short alignments, as shown in figure [1](#F1){ref-type="fig"}. Since we were interested in non-coding features of the genome, we ignored all regions where an alignment overlaps an annotated gene structure. This removed 20.8% of aligned bases. It is possible that some genes, and especially pseudogenes, have been missed by the annotation process, so we also removed portions covered by ab initiogene predictions from the Genscan program \[[@B9]\]. This eliminated an additional 4.3% of aligned bases. Finally, repetitive sequence elements annotated by the programs RepeatMasker \[[@B10]\] and trf \[[@B11]\] (5.9%) were removed from the working set. The remainder of the aligned regions were split into non-overlapping 200 base windows, ignoring any portions less than 200 bases. This gave a set of 13925 sequences which are well-conserved between human and mouse -- and therefore likely to be functional -- but which are very unlikely to be part of the protein-coding repertoire. These formed the positive training set for our machine learning strategy. A negative training set of equal size was prepared by picking 200-base windows at random from the non-coding, non-repetitive portions of chromosome 6, using the same criteria to define repeats and coding sequence. While it is probable that this set also included some functional sequences, we would expect them to be represented at a substantially lower level than in the conserved set. These two sets of sequence were presented to the Eponine Windowed Sequence machine learning system, as described in the methods section. Randomly chosen 5-base words were used as seed motifs, and three independent training runs were performed, each for 2000 cycles. The set of motifs used in model 1 is shown in table [1](#T1){ref-type="table"}. While the exact set of motifs used in the model varied somewhat from run to run, testing pairs of models on non-overlapping windows from a 1 Mb region of human chromosome 22 and plotting the scores showed that the model outputs were highly correlated (e.g. figure [2](#F2){ref-type="fig"}). We calculated the Pearson correlation coefficient for all pairs, and in all cases this was greater than 0.96. From this strong correlation, we concluded that any variations in the model were simply the result of the trainer picking one representative from a group of motifs which provide similar information. We scanned genomic sequences using these models at a range of thresholds, and examined the results on the Ensembl genome browser \[[@B12]\] using a Distributed Annotation System \[[@B13]\] server. Visual inspection showed that many of the highest-scoring regions were localised near the start of genes. This prompted us to look at the distribution of high-scoring sequences with respect to the starts of a set of well-annotated genes. We considered the GD\_mRNA genes from version 2.3 of the human chromosome 22 annotation. These are confidently annotated genes with experimental evidence as described in \[[@B14]\], which confirms at least the approximate location of the ends of the transcripts, and are independent from the chromosome 6 training data. Figure [3](#F3){ref-type="fig"} shows the density of predictions with EWS scores ≥ 0.90 relative to the annotated 5\' ends of these genes. This shows a strong peak of predictions close to the annotated starts, demonstrating that the model is predicting some sequences commonly located around the transcription start site of genes. Combining this observation with the fact that the model was trained from conserved (and therefore presumed functional) sequences, we believe that it is detecting signals found in the promoter regions of genes. Evaluation of promoter-prediction methods on a large scale is a difficult exercise, since there are no large pieces of genomic sequence for which we can be certain we know the complete set of transcribed regions, and even in the case of well-known genes we often do not know the precise location at which transcription begins. In \[[@B6]\], we developed a pseudochromosome, derived from release 2.3 of the chromosome 22 annotation. As described above, this includes a subset of 284 experimentally verified gene structures. The pseudochromosome was constructed to include these genes while omitting all other annotated genes (which could be substantially truncated). We considered predictions (groups of one or more overlapping windows which all have scores greater than some chosen threshold) to be correct if they lie withing 2 kb of an annotated gene start, and false otherwise. Plotting accuracy (proportions of predictions which are correct) against coverage (proportion of transcript starts which are detected by one of the correct predictions) gives a Receiver Operating Characteristic (ROC) curve. Using this criterion, a totally random set of predictions would be given an accuracy of around 0.07. ROC curves are plotted for the three independently trained models in figure [4](#F4){ref-type="fig"}. Firstly, this shows that predictive performance for all three models is rather similar. It also shows that they can function as accurate promoter predictors, with accuracy rising to a plateau of around 0.7, much higher than expected for random predictions. We picked model 1 for further study. Using a score threshold of 0.91, this gives an accuracy of 0.68 and a coverage of 0.31. We compared the set of genes correctly detected by this model to two other methods: firstly, the EponineTSS predictor described in \[[@B6]\], and secondly, the published results from the PromoterInspector program \[[@B15]\]. PromoterInspector results were mapped to pseudochromosome coordinates using the procedure described in \[[@B6]\]. Figure [5](#F5){ref-type="fig"} shows how the set of promoters detected by these three distinct methods overlaps. There are clearly strong correlations between all three methods. In particular, at this threshold the EWS homology model detects 98 promoters which were found by at least one of the other methods, but only 4 novel promoters. We investigated the robustness of the signal learned by this process by retraining models with a variety of seed word sizes, from 2 to 6 bases. During training, motifs can be trimmed to lengths shorter than that of the seed words (down to a minimum of 2 bases) but can never grow longer than the seed word size. When evaluated on the pseudochromosome, the resulting models always showed a preference for regions around gene starts, regardless of word length, as shown in figure [6](#F6){ref-type="fig"}. However, the accuracy was reduced when using short seed words -- particularly words of length of 2. The best accuracy was seen for a seed word length of 5, and decreased somewhat for words of length 6. This suggests that a large fraction (but not all) of the information learned by these models can be encoded in dinucleotide frequencies. It is well known that many transcription start sites are close to regions of relatively high CpG dinucleotide composition (CpG islands) \[[@B16]\]. To investigate the contribution that CpG dinucleotides make to our models, we deleted all CpG dinucleotides from the training data, then re-evaluated the resulting models on the pseudochromsome (also with CpG dinucleotides removed), as shown in figure [7](#F7){ref-type="fig"}. Perhaps not surprisingly, dinucleotide models now show very little tendency to detect gene starts. However, as the word size increases, the preference for gene starts gradually increases, until a seed size of 6 gives an accuracy comparable to that see when CpG dinucleotides are included, although the maximum coverage before accuracy begins to drop rapidly is somewhat lower. Broadly similar results are seen if CpG dinucleotides are randomly replaced with other dinucleotides. Conclusions =========== We have shown here that, when presented with a set of non-coding sequences which are strongly conserved between human and mouse, a simple motif-oriented machine learning system consistently builds models which are able to detect a substantial fraction of human promoter regions with good accuracy. This strongly suggests that this promoter signal represents the most widely used motif-based signal in functional non-coding sequence. While the model learned here can clearly be applied for the purpose of genome-wide promoter annotation, in practise existing methods offer better coverage and (in the case of the EponineTSS predictor) predictions for the precise location of the transcription start site. It is interesting that the promoter model learned by this technique detected substantially the same set of promoters as found by the EponineTSS and PromoterInspector methods. It has previously been remarked that these two methods detect similar sets \[[@B6]\], but this could perhaps be explained by the fact that both methods were initially derived from similar sets of known promoter sequences (in both cases, training data was extracted from the EPD database \[[@B17]\]. In the case of the homology models described here, there is no connection with EPD, or any similar set of known promoters: the training data was picked purely on the basis of its high similarity to corresponding portions of the mouse genome. These results therefore support the alternate view that there is a particular \'easily detected\' subclass of promoter sequences. One distinct group of promoters, which previous results show may correspond to this easily detected family, is the set of promoters associated with CpG islands \[[@B16]\]. However, while a number of the motifs listed in table [1](#T1){ref-type="table"} are G/C rich and/or contain the CpG dinucleotide, by no means all of the motifs match this description, and indeed one motif containing CpG has a negative weight in the linear model -- its presence in a sequence will reduce the model\'s output score -- while some A/T rich motifs have positive weights. We therefore believe that the signals detected here are significantly more complex than a simple over-representation of CpG dinucleotides. Experiments with smaller seed-word sizes support this assumption: while dinucleotide-based models were also able to predict promoter regions, the accuracy was lower than for models including longer motifs. Finally, we show that while the predictive capacity of dinucleotide models is largely eliminated once CpG dinucleotides are removed from the sequence, models including longer words are still able to make correct promoter predictions in many cases. So while CpG dinucleotides are an important contribution to the promoter signal, they are clearly not the only component. Methods ======= Genomic sequence and annotation ------------------------------- Human genome sequence release NCBI33 and mouse genome release NCBIM30 were extracted from Ensembl databases \[[@B12]\], which also contained gene predictions from Genscan \[[@B9]\] and repeat data from RepeatMasker \[[@B10]\] and trf \[[@B11]\]. Curated annotation of gene structures on human chromosome 6 was obtained from the Vega database \[[@B18]\]. Vega and Ensembl data was extracted directly from the SQL databases using the BioJava toolkit with biojava-ensembl extensions \[[@B19]\]. Genome alignments ----------------- Human-mouse genome alignments were generated by the blastz alignment program. These were subsequently re-scored and filtered to give a \'tight\' set of high-confidence alignments, as described in \[[@B4]\]. We downloaded the tight alignment set from the UCSC genome website \[[@B20]\]. Pseudochromosome for testing promoter-finding methods ----------------------------------------------------- A 16.3 Mb pseudochromosome sequence was produced based on version 2.3 of the curated annotation for human chromosome 22. This includes all the experimentally-validated gene structures and their upstream regions, while omitting regions containing genes that are predicted but not fully verified. In the case of a pair of divergent genes where one has been verified and the second has not, their shared upstream region was cut at the midpoint. More information about pseudochromosome construction is given in \[[@B6]\]. Eponine Windowed Sequence learning ---------------------------------- The Eponine Windowed Sequence (EWS) model is designed by analogy to the Eponine Anchored Sequence model first described in \[[@B6]\], but rather than targeting individual points in the sequence, it is designed to classify small regions or windows of a sequence, based purely on their own sequence content. The EWS model uses the Relevance Vector Machine \[[@B5]\] algorithm to drive the training process. Relevance Vector Machines solve classification and regression problems by building Generalised Linear Models (GLMs) as weighted sums of a \"working set\" of basis functions. During the training process, those basis functions which are not informative are given weights close to zero and eventually discarded from the working set. To explore very large sets of possible basis functions, it is possible to add extra basis functions during the course of the training process \[[@B6]\]. The \"sensors\" of the EWS model are DNA position-weight matrices \[[@B7]\], which make convenient models of short sequence motifs. When using weight matrices to analyse sequence windows, we sum the weight matrix probability scores for all possible positions within the sequence. Normalising for the length of the sequence being inspected and the size of the PWM, the basis functions of the model take the form: ![](1471-2105-5-131-i1.gif) where W(s) is the probability that sequence swas emitted by weight matrix W, \\|S\\| is the sequence length, \\|W\\| is the weight matrix length, and ![](1471-2105-5-131-i2.gif) denotes a subsequence from ito j. An initial set of basis functions is proposed by taking all possible DNA motifs of a specified length (typically 5) and generating weight matrices which preferentially recognise these motifs. As the relevance vector machine trainer removes non-informative basis functions from the working set, they are replaced by applying one of the following sampling strategies to a basis function picked randomly from the working set: • Generate a new weight matrix in which each column is a sample from a Dirichlet distribution with its mode equal to the weights in the corresponding column of the parent weight matrix. • Generate a new weight matrix one column shorter than the parent by removing either the first of the last column. By using these sampling rules, the trainer is able to explore motif space. The process of generating candidate motifs using these rules then selecting the most informative using the RVM can be seen as a form of genetic algorithm. Authors\' contributions ======================= TD and TH conceived and designed this study, and analysed results. TD implemented the Eponine machine learning system and drafted the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ Chromosome 22 annotation data version 2.3 were produced by the Chromosome 22 Annotation Group at the Sanger Institute and were obtained from the World Wide Web at http://www.sanger.ac.uk/HGP/Chr22 (Dunham et al. unpublished data). TD would like to thank the Wellcome Trust for funding. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Blastz alignments between human chromosome 6 and the mouse genome.Histogram showing number of alignments covering human sequences of various lengths. ::: ![](1471-2105-5-131-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Correlation of model scores.Scatter plot showing the scores of EWS models 1 and 2 on a set of human sequences. ::: ![](1471-2105-5-131-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Localisation of predictions.Density of predictions from one of the homology models around known gene starts on human chromosome 22 ::: ![](1471-2105-5-131-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Accuracy and coverage of TSS prediction.Plots of Accuracy vs. coverage at a range of score thresholds (ROC curves) for three independently trained homology models ::: ![](1471-2105-5-131-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Comparison of TSS prediction methods.Sets of pseudochromosome promoters correctly predicted by three different prediction methods: EponineTSS \[6\] with a score threshold of 0.999, PromoterInspector (labelled \"Pro\'spector\"), and the homology-EWS model 1 with a score threshold of 0.91 (\"Homol\_1\"). ::: ![](1471-2105-5-131-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Effect of seed-word size of learning.Accuracy vs. coverage plots for models trained using seed word lengths of 2 to 6 bases. ::: ![](1471-2105-5-131-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Effect of excluding CpG dinucleotides.Accuracy vs. coverage plots for models trained using a range of seed-word sizes, with all CpG dinucleotides removed from both training and test data. ::: ![](1471-2105-5-131-7) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Motifs used in EWS homology model 1. The entries in this table show consensus sequences of the weight matrices used in the model (note that it is possible for two distinct weight matrices to have the same consensus sequence). Motifs are listed in both forwards and reverse-complement orientation, and the two sections of the table indicate whether that motif is given a positive or negative weight in the learned linear model. ::: Postive Negative --------- ---------- ------- ------- gtca tgac tacgt acgta tattg caata gggca tgccc tgcca tggca gtca tgac ggca tgcc acaat attgt tacgt acgta gggc gcccc gtact agtac tact agta taac gtta cctcc ggagg ttt aaa ggca tgcc acaat attgt tattg caata caatt aattg tattg caata cagc gctg aaatt aattt cag ctg caat attg cggat atccg gtat atac aaatt aattt ccagg cctgg gctcg cgagc catg catg ggc gcc act agt taagg cctta aaaaa ttttt :::	PubMed Central	2024-06-05T03:55:48.607093	2004-9-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523850/", "journal": "BMC Bioinformatics. 2004 Sep 15; 5:131", "authors": [ { "first": "Thomas A", "last": "Down" }, { "first": "Tim JP", "last": "Hubbard" } ] }
PMC523851	Background ========== Aging is associated with a deterioration of cognitive function including a decrease in the ability to process and store new information \[[@B1]\]. Processes that might negatively affect cognitive function during aging are manifold. Among them, the cerebral vascular system has a major impact on brain function. Craigie first described that the density of cerebral microvessels may correlate with functional activity \[[@B2]\]. However, morphological studies of the microvessels have been inconsistent up to now. Meier-Ruge and coworkers reported an increase in capillary density in older individuals \[[@B3]\], whereas others had shown a reduction \[[@B4]\]. One important reason for the different findings in microvascular changes might be the heterogeneity of the examined brain regions between, but also within the studies. Nevertheless changes in the cerebral microvessels have a major impact on secondary pathophysiological changes, like reduced cerebral blood flow (CBF) \[[@B5],[@B6]\], and a decrease in doppler sonographic blood flow velocity in old age \[[@B7]\]. In addition, microvascular alterations are responsible for reduced cerebral metabolic rates for oxygen and cerebral glucose utilization, which is observed with increasing age \[[@B8],[@B9]\] leading to an impaired transport of nutrients which in turn impairs neuronal function. Additional factors such as chronic hypertension may accelerate the progression of age-related capillary changes \[[@B10]\]. Although changes of cerebral vessels with aging and hypertension have been reported, there are presently no consistent data on the microvascular basal lamina. The aim of our study was an evaluation of age-related changes on cerebral microvessels and on the vascular extracellular matrix in humans and the possible impact of chronic hypertension using several different immunohistochemical methods for the detection of collagen type IV. Results ======= Determination of collagen type IV in microvessels by immunohistochemistry ------------------------------------------------------------------------- Old patients (OP) showed 12.0 +/- 0.6 vessels per microscopic field, and young patients (YP) exhibited 15.1 ± 1.2 vessels per microscopic field (p = 0.02). The total area of collagen type IV positive vessels did not differ between OP and YP (11.0 ± 0.5 vs. 11.4 ± 0.6, n.s.). Therefore the calculated content of collagen type IV per vessel was higher in OP than in YP (1.01 ± 0.06 vs. 0.76 ± 0.05, p = 0.01). No differences were observed between old non hypertensive patients (ONHP) and old hypertensive patients (OHP) (vessels per microscopic field: 11.3 ± 1.1 in ONHP vs. 12.7 ± 0.7 in OHP, n.s.; area of collagen type IV: 12.1 ± 0.8 in ONHP vs. 10.6 ± 0.9 in OHP, n.s.; calculated content of collagen type IV per vessel: 1.11 ± 0.1 in ONHP vs. 0.91 ± 0.07 in OHP, n.s.). Thickness of vessel wall and lumen ---------------------------------- Microvessel wall thickness, inner lumen and ratio lumen/wall thickness were statistically different in OP vs. YP. The vessel wall was thicker in old than in young patients (3.14 ± 0.10 μm vs. 1.62 ± 0.06 μm, p \< 0.0001). In addition the vessel lumen was reduced in the old group compared to the young group (4.00 ± 0.14 μm vs. 5.24 ± 0.13 μm, p \< 0.0001, Figure [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}). Therefore the ratio between the thickness of the vessel lumen and the vessel wall was lowered in the OP group compared to YP (1.29 ± 0.05 vs. 3.29 ± 0.15, p \< 0.0001). The comparison of the same parameters in ONHP vs. OHP showed no significant distinctions (thickness of vessel wall: 3.13 ± 0.12 μm in ONHP vs. 3.16 ± 0.18 μm in OHP, n.s.; vessel lumen: 4.00 ± 0.26 μm in ONHP vs. 4.00 ± 0.14 μm in OHP, n.s.; ratio: 1.28 ± 0.07 in ONHP vs. 1.30 ± 0.08 in OHP, n.s.) Relative collagen type IV intensity in microvessels determined by CLSM ---------------------------------------------------------------------- Analysis of the relative intensity of collagen type-IV positive vessels was performed by confocal laser scanning microscopy. The relative content of collagen type IV in the microvessel wall was higher in OP than in YP (233.1 ± 4.5 vs. 167.8 ± 10.6, p \< 0.0001). Again, the intensity in non-hypertensive old persons showed no difference compared to hypertensive old patients (230.7 ± 4.2 vs. 234.3 ± 6.7, n.s., Figure [3](#F3){ref-type="fig"}). Discussion ========== In the present study, we investigated age- and hypertension-related alterations in the vascular extracellular matrix and the basal lamina in human brains. Our main finding is the age-related change of the basal lamina component collagen type IV. In old as compared to young individuals we found a significant decrease of the vessel number containing collagen type IV, a thickening of the vessel wall and narrowing of the vessel lumen and an increase in collagen type IV content per vessel. These changes in extracellular matrix proteins were demonstrated by immunohistochemistry as well as by confocal laser scanning microscopy. However, we were unable to establish significant changes of the microvessels in hypertensive old persons compared to normotensive old patients. These results are in good accordance with previous studies that have shown basal lamina thickening in experimental studies \[[@B10]-[@B12]\]. Research on basal lamina changes in humans has mostly been focused on their relationship to neurodegeneration. In Alzheimer\'s patients Kalaria and coworkers found a 55% increase of collagen type IV content in cerebral microvessels in comparison to age-matched controls \[[@B13]\]. Farkas and colleagues described collagen accumulation in the basal lamina of Parkinson\'s disease \[[@B14]\]. Data about changes in the cerebral microvasculature in normal aging, however, are scarce. One study examined cerebral microvessels in human neocortex, yet failed to demonstrate an age-related thickening of the basal lamina \[[@B15]\], however one study was able to show a decrease of microvascular density by age in the hypothalamus \[[@B4]\], interestingly also this study failed to show hypertension related changes. One explanation for the decreased microvascular density in the putamen might be a reduced neoangiogenesis in the aging human brain, this hypothesis should be examined in further studies. One reason for the inconclusive findings in human cerebral microvessels might be the local origin of the samples, as up to now rarefaction was only found in the deep grey matter. The microvasculature is organized differently in the basal ganglia than in the cortex. The basal ganglia microvasculature has the geometry of a tree-like vascular bed, in contrast the cortical microvascular networks are rather organized like a grid structure \[[@B16]\]. Due to this reason the deep grey matter might be more vulnerable to metabolic changes in age, with reduced capabilities for neoangiogenesis, as this might compensatory occur in cortical areas. Therefore differential patterns of microvascular changes within the brain might occur, resulting in controversial results of microvascular density in the aging brain. Another explanation seems to be a methodological one. As the region of interest in putaminal sections can be exactly defined from section to section, analyses of cortical areas might be more problematic resulting in an imprecise definition of comparable areas \[[@B15]\]. The reasons for the alterations of vascular extracellular matrix proteins in aging are largely unknown. Not only age is associated with an increase of extracellular matrix proteins, as these changes were observed in several diseases, like hypertension, brain tumors \[[@B17]\], HIV-encephalopathy \[[@B18]\], Alzheimer\'s \[[@B13]\] or Parkinson\'s disease \[[@B14]\]. One explanation for the extracellular matrix accumulation might be the reduction of the proteolytic systems activity. The matrix metalloproteinases (MMP) and the natural tissue endogenous inhibitors (TIMP), as well as the plasminogen/plasmin system are involved in the regulation of ECM metabolism \[[@B19],[@B20]\], and changes in these proteases activity may contribute to vascular remodelling in age by modulating the extracellular matrix components. Therefore a reduction of these proteases might result in a decreased turnover of the basal lamina with a consecutive increase of these components. The reduction of MMP activity by age was recently shown in an experimental study \[[@B21]\] and in humans by antihypertensive treatment \[[@B22]\], however studies about the role of these proteases in the aging human brain are lacking and the impact has to be evaluated in further studies. The strength of our study is the combination of several different methods for the detection of basal lamina changes. Even if the study by Abernethy and colleagues \[[@B4]\] was the first one, that showed age related changes in the deep grey matter, their study has some limitations. First they used a nonspecific staining technique with alkaline phosphatase, and second no detailed morphometric analysis of the vessels or changes of the basal lamina were performed. On the other hand, previously published studies most often used only one method for the determination of microvascular changes. For example the analyses of the vessel wall and lumen was widely used as the only parameter. \[[@B23]-[@B25]\] Unquestionably this method has a subjective approach and therefore we see our results of the increased vessel wall/lumen ratio in aging as a confirmation of previous results. But in contrast to these studies our approach employed an additional variety of complementary methods to examine the increase of basal membrane components: number of vessels, total stained area of collagen type IV, relative collagen type IV content per vessel and relative immunofluorescence by CLSM, all indicating into the same direction. Some methodological issues have to be discussed, starting with the unexpected lack of a difference between old normotensive and old hypertensive patients. One explanation might be due to silent hypertension in the ONHP group, as well as treated hypertension in the OHP group. We tried to minimize this problem due to carefully study of the case records in all patients. In addition in our study hypertension was defined as a history of hypertension, rather than the actual blood pressure in hospital, as these patients with severe diseases might not had representative blood pressure values in their last days or weeks of life than decades before. On the other hand another study failed to show the impact of hypertension on microvascular densitiy in a neuropathological study of the human hypothalamus \[[@B4]\]. This unexpected lack of hypertension related changes in the microvessels should be regarded as preliminary, as neither the duration of hypertension nor the effectiveness of treatment was considered. Conclusions =========== The present data show age-related changes of the cerebral microvessels in sections of human putamen for the first time. Due to the accumulation of collagen, microvessels thicken and show a reduction in their lumen. Besides this, the number of vessels decreases. These findings might represent a precondition for the development of vascular cognitive impairment. Methods ======= The study was performed on 27 post-mortem human brain samples from the putamen, which were taken from autopsy. The clinical diagnoses were confirmed by routine pathology and are shown in the table. Two groups of subjects were compared: first young patients, all without a history of hypertension (YP; n = 11, mean age 38.8 ± 6.8 years), and old patients (OP, n = 16, mean age 73.9 ± 4.1 years). The old patients were divided into two subgroups, those without a history of hypertension (ONHP, n = 8, mean age 73.1 ± 4.9 years), and those with a history of hypertension (OHP, n = 8, mean age 74.6 ± 3.4 years). There were no significant differences in age between OHP and ONHP and in sex between YP and OP, as well as between ONHP and OHP (see Table [1](#T1){ref-type="table"}). The putamen either of the right or left side were removed completely and fixated in paraffin. We chose the putamen region, as it is easily to define and vascular changes and strokes are predominantly located in this area. The blocks were cut cross sectional in the same anterior-posterior direction resulting in axial sections with a thickness of 10 μm. The sections were deparaffinized and immersed at 37°C in 0.4% Pepsin (Sigma, Germany) in 0.01 N HCl for one hour. Collagen IV-positive vessels were stained with a monoclonal mouse anti-collagen-IV antibody (Sigma, Germany). Each section was incubated with 150 μl of the primary antibody solution (at a concentration of 1:200) for two hours at 37°C followed by incubation with biotinylated secondary antibody against mouse IgG for 30 minutes at 37°C (Vector Laboratories). Vectastain ABC reagent was added for 30 minutes at 37°C. Chromogen (AEC Kit Biomeda Corp.) was used to develop the peroxidase signal. Negative and positive controls were routinely performed in each staining experiment. The same procedure was used for immunofluorescence staining. Instead of using the Vectastain ABC kit containing avidin, avidin marked FITC (Dianova, Hamburg, Germany) was added for 30 minutes at a dilution of 1:100. The number of peroxidase stained vessels was determined with the aid of a computerized video imaging system at a magnification of ×100 (Optimas Version 6.5 from Media Cybernetics, Silver Spring, USA). Only vessels smaller than 30 μm were included. Total area of collagen IV positive vessels in the sections was analyzed using the same system. Results are presented in arbitrary units. To obtain the relative amount of collagen type IV per vessel the area of collagen type IV was divided by the number of stained vessels per microscopic field (\[collagen type IV/microscopic field\]/\[vessels/microscopic field\]) The size of the observed microscopic field was 150 × 200 μm. To estimate microvessel hypertrophy, the ratio between the diameter of vessel lumen and vessel wall, respectively, was calculated semiquantitatively with the help of a second computerized video imaging system (Medmo, Homburg, Germany). Twenty entire cross-sectional microvessels from the putamen were randomly selected at a magnification of ×400. To calculate the wall to lumen ratio, average distances of vessel wall and vessel lumen were selected. Fluorescence intensity measurements of microvessel-associated FITC anti-mouse IgG against the anti-collagen antibody were performed with confocal laser scanning microscopy (CLSM, Leica, Heidelberg, Germany). All measurements were performed with the same pinhole size, brightness and contrast, zoom, and laser time. Each vessel was scanned in the zplane (10 scans per 1 μm), and a summed image was calculated. Also, a summed image was obtained from the background area to normalize the local intensity to the background. The normalized intensity is expressed as mean ± SEM for each microvessel using a scale from 0 to 255 arbitrary units (U). The technique was adopted from Hamann et al \[[@B26]\]. Twenty randomly selected microvessels each of 7.5 to 30 μm in diameter of the putamen were measured in each specimen. Statistical analysis -------------------- Data are presented as mean +/- standard error of mean. Statistical evaluations were performed using t-test. Authors\' contributions ======================= OU carried out the immunohistochemical experiments, ML performed the statistical analysis and drafted the manuscript. JH participated in the design of the study and collected the brain specimens. AD participated in the study design. GFH supervised the thesis, and participated in its design and coordination. Acknowledgements ================ The technical assistance of Mrs. Nathalie Wunderlich and Mrs. Gabriele Jaeger is gratefully acknowledged. We thank Mrs. Judy Benson for copyediting the manuscript. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Analysis of the thickness of the vessel wall and the inner diameter of the vessel lumen in YP and OP.Figure A indicates YP, Figure B indicates OP. ::: ![](1471-2202-5-37-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Analysis of the thickness of the vessel wall and the inner diameter of the vessel lumen.Black boxes indicate YP. OP were divided into ONHP (grey boxes), and OHP (white boxes). The YP group revealed a significant thinner vessel wall and a larger vessel lumen than OP (p \< 0.0001). No difference was observed between ONHP and OHP. ::: ![](1471-2202-5-37-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Analysis of the relative collagen type IV content in the microvessels by CLSM.The difference was significant between YP and OP (not shown), but not between ONHP and OHP. ::: ![](1471-2202-5-37-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Charactistics and cause of death ::: Young patients Old non-hypertensive pat. Old hypertensive patients -------------------- ------------------------------- ------------------------------- ----- -------- ----- -------------------- ----- -------- ----- --------------------- ----- No Age Cause of death Sex No Age Cause of death Sex No Age Cause of death Sex 1 40 Bronchial carcinoma f 12 68 Lung embolism f 20 81 Gastric hemorrhage f 2 37 Malignant melanoma f 13 67 Lymphoma F 21 70 Pneumonia f 3 27 Lung embolism f 14 82 Mamma-carcinoma f 22 77 Peritonitis f 4 42 Leukemia f 15 71 Lymphoma F 23 74 Resp. insufficiency m 5 34 HIV f 16 77 Liver cirrhosis f 24 72 Bypass surgery m 6 40 Bronchial carcinoma f 17 72 Heart-lung insuff. m 25 74 Hepatorenal syndr. m 7 30 Aplastic anemia f 18 74 Sigmoid-carcinoma m 26 76 Sigmoid-carcinoma f 8 47 Leukemia m 19 74 Stomach-carcinoma m 27 73 Plasmacytoma f 9 44 Aortic aneurysm m 10 49 Hodgkin\' disease m 11 37 Leukemia f :::	PubMed Central	2024-06-05T03:55:48.609060	2004-9-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523851/", "journal": "BMC Neurosci. 2004 Sep 24; 5:37", "authors": [ { "first": "Olga", "last": "Uspenskaia" }, { "first": "Martin", "last": "Liebetrau" }, { "first": "Jochen", "last": "Herms" }, { "first": "Adrian", "last": "Danek" }, { "first": "Gerhard F", "last": "Hamann" } ] }
PMC523852	Background ========== Genetic analysis of inherited deafness is a powerful tool for discovering the molecular mechanisms that control the development, function and maintenance of the auditory system. Linkage analysis of deafness segregating in large families has resulted in the mapping of more than 55 non-overlapping loci and the discovery of over 30 genes that are essential for hearing \[[@B1]\]. Given this extensive genetic heterogeneity, extended families from endogamous populations are ideally suited for identifying novel deafness genes and for genotype-phenotype studies. DFNB8/B10, an autosomal recessive deafness locus, was independently mapped in two consanguineous families from Palestine and Pakistan \[[@B2],[@B3]\]. Haplotype and gene sequence analyses of individuals in these two families led to the identification of mutations in a gene encoding a serine protease, TMPRSS3 \[[@B4],[@B5]\]. TMPRSS3 belongs to a subfamily of type II transmembrane serine proteases, which also includes TMPRSS1, TMPRSS2, TMPRSS4 and TMPRSS5 \[[@B6]\]. The TMPRSS3gene, spanning approximately 24 kb on chromosome 21, contains thirteen reported exons \[[@B4]\]. In humans there are alternatively spliced transcripts (TMPRSS3 a, b, cand d), encoding predicted polypeptides of 454, 327, 327 and 344 amino acids, respectively \[[@B4]\]. Here, we report the identification of a fifth isoform, TMPRSS3e, which encodes 538 amino acid residues, including a signal peptide, and is expressed in many human tissues. We previously reported pathogenic TMPRSS3mutations in four out of a total of 159 Pakistani families segregating profound congenital recessive deafness \[[@B5]\]. To determine the contribution of TMPRSS3mutations to recessive deafness in Pakistan, we screened an additional 290 Pakistani families for linkage to the DFNB8/B10 locus. We also screened TMPRSS3for mutations in a large kindred from Newfoundland, Canada segregating hearing loss linked to markers for DFNB8/B10 and surprisingly found two different mutant alleles. Methods ======= Family ascertainment -------------------- Institutional review board approvals (OH93-N-016 and OH95-DC-N-050) were obtained for this study from the National Institutes of Health, USA, the Centre of Excellence in Molecular Biology, Lahore, Pakistan and the Newfoundland and Labrador Medical Genetics Program, Health Sciences Centre, St. John\'s. Participating individuals gave written informed consent. Medical histories indicated that all four Pakistani families segregate congenital, profound, non-syndromic sensorineural hearing loss. Pure tone air and bone conduction audiometry was performed on affected and unaffected individuals of the family from Newfoundland. Pneumatic otoscopy was used to confirm or deny cases in which conductive hearing loss was suspect. All testing was done in non-sound-attenuated exam room. Prior to testing, a biologic calibration was done in the room used for testing with a listener having normal hearing thresholds (LS). Possible effects of ambient noise on thresholds were then taken into account and only those frequencies unaffected were included. DNA was extracted from venous blood samples from the participating individuals. Linkage and sequence analyses ----------------------------- DNA samples were PCR amplified using fluorescently labeled primers surrounding microsatellite repeats at known DFNB loci and analyzed on an ABI 377 DNA sequencer. Genotypes were determined using Genescan and Genotyper software (PE Biosystems). To detect TMPRSS3mutations in families segregating DFNB8/B10, the sequence of each exon of TMPRSS3was evaluated in two affected subjects from each family as described previously \[[@B5]\]. 5\' RACE and cloning of TMPRSS3e ---------------------------------- Sequence analysis of the subcloned PCR product amplified from human retina GETLarge full length cDNA (Genemed Biotechnologies) using primers 5\'GGGTTGCTTCAAATGGCTTACTAGATCC3\' and 5\'CATTTTCCCCCATGGTGACTATTTCAG revealed an additional 385 bp of transcribed sequence downstream of TMPRSS3aexon 1. 5\' RACE PCR using human retina Marathon-Ready cDNA as template (Clontech) with primers 5\'CAGACCAATGGCCAGTGCTAATATC3\' and the AP1 primer was performed using the following thermal cycling conditions: 94°C for 1 min, 25 cycles of 94°C for 30 s and 72°C for 5 min. Five microliters of amplified PCR product was used for the second round of amplification with nested primer 5\'TTTTCAAATCATCAAGGCCAAAAAG3\' and the AP2 primer. Five microliters of each sample were analyzed on a 1.2% agarose/ethidium bromide gel, extracted using a QIAquick gel extraction kit (Qiagen) and cloned into E. coliusing pGEM^®^-T Easy vector (Promega). DNA was purified from minipreps using QIAquick miniprep kit (Qiagen) and inserts were sequenced using T7 and SP6 primers. To amplify the full length TMPRSS3eisoform, 5\'ATGGTGAGTAAAATGGGTGTGAGGA3\' and 5\'CTTGGAAGTAGAAAGGGTGGGTTTG3\' primers and LA-Taq polymerase (Pan Vera) were used as recommended by the manufacturers. Multiple-tissue cDNA panel analysis ----------------------------------- Expression of TMPRSS3ewas evaluated using a Human Tissue cDNA panel (Clontech). TMPRSS3ecDNA was amplified with primers 5\'CCAGAAATGGTGAGTAAAATGG3\' and 5\'AGCAACAGCATCTGCATCTGGT3\' using 20 pg of each cDNA as template following the manufacturer\'s recommended protocol (Clontech). PCR amplified products were sub-cloned into pGEM-T Easy vector (Promega) and sequenced. Results ======= In four out of 290 newly investigated Pakistani families, nonsyndromic congenital deafness was found to co-segregate with DFNB8/B10 linked haplotypes (Figure [2](#F2){ref-type="fig"}). Mutational analysis of DNA samples from the affected individuals from these families (Table [1](#T1){ref-type="table"}) revealed three previously reported mutations (207delC, C407R, C194F). In our families segregating 207delC there is a common disease associated haplotype \[[@B5]\]. The second mutation, C407R, was found previously in affected individuals of two Pakistani families and in a heterozygous state in one person of Indian descent. The same C407R ancestral haplotype was also found in family PKDF040 \[[@B5]\]. The third mutation, C194F, was reported in one Pakistani family \[[@B5]\], which has the same haplotype as family PKDF064 (Figure [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Pedigrees of Pakistani families.Four families with nonsyndromic recessive deafness from Pakistan. ::: ![](1471-2350-5-24-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Mutant alleles of TMPRSS3found in this study ::: ---------- ------------ -------- ------------------------- ----------------- Family Mutation Domain Amino acid substitution First reference Family B IVS8+8insT SP Frameshift This study Family B 207delC LDLRA Frameshift \[9\] PKDF003 207delC LDLRA Frameshift \[9\] PKDF040 1219T\>C SP C407R \[5\] PKDF064 581G\>T SRCR C194F \[5\] PKDF311 207delC LDLRA Frameshift \[9\] ---------- ------------ -------- ------------------------- ----------------- SP; Serine Protease, LDLRA; Low Density Lipoprotein Receptor A, SRCR; Scavenger-Receptor Cysteine-Rich ::: After completing a genome wide screen we found linkage of deafness to markers for DFNB8/B10 in one branch of a large six-generation pedigree (Figure [1](#F1){ref-type="fig"}) from Labrador \[[@B7]\], a Canadian island whose small population lives in villages isolated by geography and weather \[[@B8]\]. Medical histories and pure tone air and bone conduction audiometry revealed nonsyndromic, pre-lingual, severe to profound, sensorineural hearing impairment in ten participating affected individuals. Linkage analysis in this family demonstrated homozygosity for DFNB8/B10 linked microsatellite markers among seven affected individuals from five different sibships (Figure [1](#F1){ref-type="fig"}). An apparent founder DFNB8/B10 associated haplotype was constructed using the available genetic information (Figure [1](#F1){ref-type="fig"}). Three hearing impaired individuals (IV:10, V:2 and VI:1) carry only a single copy of the DFNB8/B10 haplotype, although their deafness is clinically indistinguishable from that of the five homozygotes. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Pedigree of Newfoundland family.There are ten hearing impaired individuals in a six generation extended family structure. Drawn below the enrolled subjects is a haplotype for STR markers around the DFNB8/B10 locus on chromosome 21. The carrier status of each person for the mutant alleles of TMPRSS3found in this family is shown. \"C\" represents 207delC, while \"T\" stands for IVS8+8insT. Individuals V:17 and V:20 are compound heterozygotes. ::: ![](1471-2350-5-24-1) ::: Sequencing of the TMPRSS3gene in affected individuals revealed two mutant alleles. The first mutation occurs in exon 4 of TMPRSS3and is a deletion of a cytosine at position 207 (207delC) of the mRNA (Figure [1](#F1){ref-type="fig"}, Table [1](#T1){ref-type="table"}). The second mutation inserts a \"T\" residue at the eighth position after the splice donor site of exon 8 (IVS8+8insT). Although the precise effect of this mutant allele is not known, Genscan predicts this results in the skipping of exon nine. Individual IV:16 is a carrier of IVS8+8insT and her two affected children are compound heterozygotes (Figure [1](#F1){ref-type="fig"}). Neither of these two mutations was found in 100 random normal control DNA samples (200 chromosomes) from Newfoundland. We previously reported a Pakistani family (PKSR7) in which deaf subjects were homozygous for the markers spanning the DFNB8/B10 region with the simulated maximal lod score of 3.8 \[[@B5]\]. However, we did not find a mutation in the known coding and non-coding exons of TMPRSS3\[[@B5]\]. One possible explanation is that family PKSR7 has a mutation in a regulatory element of TMPRSS3. Alternatively, there might be additional uncharacterized exons of TMPRSS3. Considering the later possibility, we searched for additional exons and alternatively spliced transcripts of TMPRSS3. Previous studies have documented four isoforms of TMPRSS3\[[@B4]\]. To probe for novel coding sequence of TMPRSS3we designed primers from the known exons to amplify smaller overlapping cDNA product(s). The amplicon generated using a forward primer from the reported non-coding exon 1 and a reverse primer from the protein coding region of exon 2 was larger than the expected size product based on the sequence of TMPRSS3isoform a\[[@B4]\]. After subcloning and sequencing this transcript, an additional 385 bp of sequence was found at the 3\' end of exon 1 \[[@B4]\]. We further confirmed the results by 5\'RACE using gene specific primers and human Marathon-Ready retina cDNA as template. Using specific primers we amplified a full-length transcript of this novel isoform, which we designated TMPRSS3e(Figure [3](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### *Mutational spectrum, structure and expression of TMPRSS3isoforms.(A) Coding and non-coding exons of all the known isoforms of TMPRSS3are shown with black and gray rectangles, respectively. The newly identified isoform ehas translation initiation codon (arrow) in exon 1, while the termination codon in exon 13 is marked with an asterisk. Shown also are predicted protein motifs encoded by the 3192 bp long mRNA of isoform e. All the known mutant alleles of TMPRSS3causing hearing loss are shown above the protein motifs. Modified and updated from Ben-Yosef et al. 2001 \[5\] (B) Nucleotide sequence of the cDNA encoding the amino terminus of TMPRSS3eand its deduced amino acid sequence. The underlined nucleotide sequence represents the region predicted by SMART to encode a signal peptide. The last ATG shown is the reported translation initiation site for isoform a\[4\]. (C) RT-PCR specific to the TMPRSS3etranscript was performed on cDNA from seven human tissues, which include retina, lung, liver, heart, pancreas, placenta and kidney as indicated. All tissues, except heart, demonstrated expression of TMPRSS3e. G3PDHwas used as a positive control. ::: ![](1471-2350-5-24-3) ::: Although the ATG codon at position 705 (Accession \# AY633572) does not meet Kozak consensus criteria, there are two reasons for believing that this is the translation initiation site. First, there are stop codons in all three reading frames upstream of this site. Secondly, the 252 bp open reading frame between ATG 705 and ATG 957 (Figure [3A](#F3){ref-type="fig"}), the previously reported initiation codon, includes a predicted signal peptide, which is expected in a secreted serine protease. Considering the ATG at position 705 as a starting codon, analysis of the 1614 bp open reading frame indicates that this larger TMPRSS3etranscript encodes a putative polypeptide of 538 amino acids, including 84 previously unknown amino acids at the amino terminus (Figure [3B](#F3){ref-type="fig"} and [3C](#F3){ref-type="fig"}). The SMART program predicts a cleavable signal sequence between residues 1 through 36 (Figure [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}). Among all the known isoforms of TMPRSS3, only isoform ereported here has a predicted signal peptide, consistent with the hypothesis that TMPRSS3encodes a secreted serine protease. The remaining 48 new amino acids do not contain other predicted structural or functional motifs (Figure [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}). Downstream of the newly identified 84 amino acids, isoform e shares a common protein sequence with isoform a, which includes a transmembrane (TM), low density lipoprotein receptor A (LDLRA), scavenger-receptor cysteine-rich (SRCR) and serine protease domains (Figure [3A](#F3){ref-type="fig"}). Through non-quantitative RT-PCR analysis with TMPRSS3especific primers and multiple human cDNAs, we found expression of this isoform in retina, lung, liver, pancreas, placenta and kidney (Figure [3C](#F3){ref-type="fig"}). Mutational screening of exon 1 and exon 2 as well as other exons of TMPRSS3e, using DNA samples from two affected individuals of family PKSR7 did not reveal a pathogenic change. This may be an indication of as yet uncharacterized additional exons of TMPRSS3, or the hearing loss phenotype in this family is due to a mutation in a regulatory element of TMPRSS3. Alternatively, it remains possible that family PKSR7 has a mutation in a gene closely linked to TMPRSS3. Discussion ========== TMPRSS3is the only protease reported thus far to be involved in nonsyndromic deafness \[[@B4]\]. In this study, we document the mutant alleles of TMPRSS3segregating in a large family from Newfoundland and in additional Pakistani families with nonsyndromic, recessive deafness. The Newfoundland family has at least two different mutated genes associated with hearing impairment. The majority of the affected individuals in this family are homozygous for 207delC (Figure [1](#F1){ref-type="fig"}), while two of the hearing impaired subjects are compound heterozygotes for 207delC and a newly described mutation of intron 8 (IVS8+8insT). The 207delC mutation appears to be wide-spread; it has been found in affected individuals of Spanish (one homozygote), Greek (one heterozygote), Newfoundland (homozygous individuals) and Pakistani (homozygous individuals, PKDF003 and PKDF311) origins, and is the most common of the mutant alleles of TMPRSS3. Although there is a predicted effect of IVS8+8insT mutation on splicing of exon 9, further studies are required to determine the precise role of IVS8+8insT on TMPRSS3transcripts. The family from Pakistan that originally defined the DFNB8 locus was reported to have childhood onset hearing impairment \[[@B3]\], while all remaining families segregating TMPRSS3mutations have pre-lingual deafness \[[@B4],[@B5],[@B9],[@B10]\]. Mutational screening revealed a splice site mutation (IVS4-6G\>A) in the original DFNB8 family, which may allow some normal splicing of TMPRSS3transcripts \[[@B4]\]. The second mutant allele (IVS8+8insT) in the Newfoundland family may also be a leaky mutation, although the two individuals who are compound heterozygotes were described as having prelingual hearing impairment. The hearing impaired individuals IV:10, V:2 and VI:1 of our Newfoundland family are carriers of 207delC and we did not find any other TMPRSS3mutation in these individuals. Individuals IV:10 (affected) and IV:12 (normal hearing sibling) show an identical chromosome 21 haplotype. Similarly the individuals VI:1 (affected) and VI:2 (normal hearing sibling) share identical haplotypes for the TMPRSS3locus. Hence, it is unlikely that the deafness of individuals IV:10 and VI:1 is due to cryptic TMPRSS3mutations (Figure [1](#F1){ref-type="fig"}). As connexin 26 mutations are the most common cause of nonsyndromic recessive deafness \[[@B11]\], we sequenced the coding exon of GJB2in affected individuals of Newfoundland family and found no mutation. We have analyzed a total of 449 Pakistani families \[\[[@B5]\] and this study\] segregating severe to profound congenital recessive deafness and found a total of eight families in which the deafness phenotype is due to mutations of TMPRSS3. Therefore, the relative contribution of TMPRSS3mutations in the deaf Pakistani population is approximately 1.8%, a significant amount considering the extensive genetic heterogeneity of deafness in this population. Of ten TMPRSS3mutations that have been reported world wide, five (IVS4-6G\>A, 207delC, R109W, C194F and C407R) are segregating in the Pakistani population \[[@B4],[@B5],[@B9],[@B10]\]. Among the ten known mutant alleles of TMPRSS3, six are present in exons common to all isoforms, including isoform \"e\" reported herein. Of the isoforms of this serine protease, TMPRSS3ehas the longest open reading frame and is the only isoform of this gene with a predicted signal sequence at the amino terminus. The identification of 84 additional amino acids, including a predicted signal sequence, may aid functional assays to determine the role of TMPRSS3in cochlear development and function. TMPRSS3isoform awas identified with a forward primer (TMa-F; see supplement table B \[[@B4]\]) in exon 2 and not in exon 1 as illustrated in figure [1](#F1){ref-type="fig"} of Scott et al. 2001 \[[@B4]\]. Therefore, it is possible that TMPRSS3ais an incomplete version of isoform eand isoform amay not exist in vivo. TMPRSS3message is expressed in supporting cells of the organ of Corti, in the stria vascularis and in the spiral ganglion cells of the cochlea \[[@B12]\]. Although the specific role of TMPRSS3in the development and maintenance of the audiosensory apparatus is still unknown, the reported mutant alleles of TMPRSS3 abolish catalytic activity of the serine protease, implying a proteolytic function during the inner ear development \[[@B12],[@B13]\]. Although the in vivosubstrate(s) of TMPRSS3have not been reported in the auditory system, TMPRSS3 is thought to regulate the activity of the epithelial amiloride sensitive sodium channel (ENaC) in vitro, which was suggested to control critical signaling pathway(s) in the inner ear and may have a role in the maintenance of the low sodium concentration of endolymph \[[@B12]\]. However, the absence of an abnormal auditory phenotype in individuals with Pseudohypoaldosteronism type I (PHA 1), which are homozygous for null alleles of ENaC subunits \[[@B14]\], suggests that the DFNB8/B10 deafness phenotype is due to aberrant proteolytic processing by TMPRSS3 of some other substrate in the inner ear. Conclusions =========== TMPRSS3mutations account for hearing loss in 1.8% (8 of 449) of Pakistani families segregating deafness as an autosomal recessive trait. We also identified two recessive mutations of TMPRSS3segregating in a six-generation extended family from Newfoundland. Our study also revealed a longer isoform of TMPRSS3with an exon encoding a signal peptide, which should help in the functional dissection of this secreted serine protease. Abbreviations ============= TMPRSS3; Transmembrane Protease, Serine 3 \[OMIM entry 605511\]. DFNB8; Deafness, childhood-onset neurosensory autosomal recessive 8 \[OMIM entry 601072\]DFNB10; Deafness, congenital neurosensory, autosomal recessive 10 \[OMIM entry 605316\] Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= ZMA performed DNA sequencing, characterized the novel TMPRSS3isoform e*and prepared the manuscript. XCL analyzed all linkage data from family B from Newfoundland and identified DFNB8/B10 segregation in this family. SDP performed DNA sequencing to determine mutant alleles of TMPRSS3 and confirmed linkage in Pakistani families. SR, KR and ZA enrolled families in Pakistan. TLY re-sampled some members of family B from Newfoundland, confirmed genotypes and sequenced normal controls from Newfoundland. SL participated in the clinical assessment of Newfoundland deaf family members. KD sequenced normal controls from Newfoundland. LM was the genetic counsellor involved in family B enrolment from Newfoundland. BP provided considerable technical assistance. LIS performed hearing tests on Newfoundland family members. EI identified Newfoundland families with inherited deafness. SR was responsible for the enrolment of 449 families from Pakistan. TBF was in charge of oversight, editing and analyses. RJM and ERW trained investigators and were involved in the experimental design and genetic analyses of all families. All authors contributed to and edited the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2350/5/24/prepub> Acknowledgements ================ We thank the families for their participation in this study, which was supported by NIDCD/NIH, Intramural Research funds Z01DC00035-06 and Z01DC00039-06. Part of this study in Pakistan was supported by Higher Education Commission, Islamabad, Pakistan and by International Centre for Genetic Engineering and Biotechnology, Trieste, Italy under Project CRP/PAK02-01 (Contract No.02/013). We thank Shaheen Khan, Tamar Ben-Yosef, Juile Schultz and Dennis Drayna for advice and comments in preparing the manuscript.	PubMed Central	2024-06-05T03:55:48.611010	2004-9-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523852/", "journal": "BMC Med Genet. 2004 Sep 24; 5:24", "authors": [ { "first": "Zubair M", "last": "Ahmed" }, { "first": "Xiaoyan", "last": "Cindy Li" }, { "first": "Shontell D", "last": "Powell" }, { "first": "Saima", "last": "Riazuddin" }, { "first": "Terry-Lynn", "last": "Young" }, { "first": "Khushnooda", "last": "Ramzan" }, { "first": "Zahoor", "last": "Ahmad" }, { "first": "Sandra", "last": "Luscombe" }, { "first": "Kiran", "last": "Dhillon" }, { "first": "Linda", "last": "MacLaren" }, { "first": "Barbara", "last": "Ploplis" }, { "first": "Lawrence I", "last": "Shotland" }, { "first": "Elizabeth", "last": "Ives" }, { "first": "Sheikh", "last": "Riazuddin" }, { "first": "Thomas B", "last": "Friedman" }, { "first": "Robert J", "last": "Morell" }, { "first": "Edward R", "last": "Wilcox" } ] }
PMC523853	Background ========== Microarray experiments produce vast amounts of data. The resulting datasets are highly complex and contain large matrices of expression measurements as well as sequence and experiment annotations that provide biological context to the data. To organize these different types of data in a way that allows intuitive exploration of the data, and provides the ability to gain important insights into relationships within a given dataset requires sophisticated visualization tools. Such visualization tools are of benefit not only to researchers analyzing and presenting or publishing their own data, but also to Model Organism Databases (MODs) for compiling and displaying microarray data for a given model organism. There are several excellent free tools available that allow an individual user to analyze their own data. These tools are either accessible on the web, or can be downloaded and used on a desktop machine. Examples include the EPCLUST \[[@B1]\], GEPAS \[[@B2],[@B3]\] and FGDP \[[@B4],[@B5]\] web-based tools and the TMEV \[[@B6],[@B7]\] desktop tool from TIGR. However, once these tools have been used, and a cluster or other group of genes has been selected, this resulting dataset needs to be made available to other people for browsing and exploration. There are a few visualization tools that allow display of such a static dataset that are available as free software tools, e.g. Michael Eisen\'s TreeView \[[@B8],[@B9]\], JavaTreeView \[[@B10]\], or the more recent MapleTree \[[@B8]\]. All of these tools are, however, desktop tools that themselves have to be downloaded and work on locally stored datasets. The impetus for the development of GeneXplorer was the desire to provide access to datasets via the Internet, without the requirement to download and install additional software. We developed GeneXplorer for use in web supplements of microarray publications whose raw data are housed within the Stanford Microarray Database (SMD) \[[@B11],[@B12]\] and for use as a tool to allow SMD users to browse their own data within SMD before publication. Using GeneXplorer, hierarchically clustered gene expression data can be interactively viewed using a web browser on any computer platform. GeneXplorer uses the widely accepted CDT file format \[[@B13]\] produced by several freely available clustering programs (e.g. \[[@B9],[@B14]\]), which between them have been downloaded several thousand times. Thus GeneXplorer should be widely usable my SMD and non-SMD users alike. Implementation ============== The application was written using object oriented Perl following the Model-View-Controller (MVC) design paradigm \[[@B15]\]. GeneXplorer consists of two classes, the data model class Microarray::CdtDataset (M), and the presentation logic class Microarray::Explorer (V). The controller, named gx, is a Perl CGI script that dispatches CGI requests to the viewer. The MVC paradigm was used because it dissociates how data are represented internally (the Model) from how they are displayed (the View), from how they are interacted with (the Controller) (see Figure [1](#F1){ref-type="fig"}.). The goal of such a separation is that by keeping consistent APIs for the components to interact with each other, each component may be modified extensively internally, with little or no effect on the other parts of the application, thus making code maintenance easier. The Microarray::CdtDataset class provides an application programming interface (API) that allows details of a particular expression cluster to be queried. In turn, instances of the Microarray::Explorer class use this API to retrieve and then display information about the dataset. The controller is a relatively simple CGI Perl script that is responsible for capturing CGI parameters and using them to first create a dataset Microarray::CdtDataset object, which is subsequently used in the instantiation of a Microarray::Explorer object. The controller then invokes the appropriate Microarray::Explorer methods, depending on where, and in which frame the user clicked. The Microarray::CdtDataset has two essential functions: during dataset creation (see below) it decomposes the data file into its constituent data parts and creates the files needed during data viewing (see below). During data viewing it provides the API for the viewer, and allows searching and retrieval of the data. Under the current model the dataset object itself is immutable. Microarray::CdtDataset was implemented as a client of the Microarray::DataMatrix module, which provides an API for accessing matrices of expression data. In the design of the classes certain compromises had to be made to accommodate the stateless client server environment in which the program operates. Specifically, to allow rapid responses, pre-generated images and correlation data are cached in a compact format on the web-server. There are two stages required to publish a microarray dataset on the web using GeneXplorer. The first stage (executed only once per dataset) involves creation of all the necessary files for GeneXplorer to use. The second stage uses these files to produce the display using the GeneXplorer web front-end. Dataset creation ---------------- Dataset creation requires a file in the Clustered Data Table (CDT) format: a simple tab delimited text file format (see \[[@B13]\] for file format details). This format was introduced with the \'Cluster\' and \'TreeView\' applications \[[@B8]\] and is widely used for microarray data. A perl script (makeMicroarrayDataset.pl) uses Microarray::CdtDataset to create the various required data files. Correlations between expression-vectors within the dataset are calculated for each pair-wise combination of vectors using the C program \'correlations\'. Correlations for each vector above the default cutoff value of 0.5 (which is configurable) are saved in a binary format that facilitates rapid searching. Depending on the version of the Perl GD module \[[@B16]\] installed on the system, either png or gif formatted images representing the cluster will be created. These images include both a 2-color image representation of the data matrix and an image representation of the experiment names. The program that creates these files is configurable, such that these images can be created using either a red/green or a yellow/blue color scheme, and in addition, the contrast of the images can be customized and set in steps of log(2) scale. The name and path of a dataset can be defined hierarchically within the file system (see Figure [2](#F2){ref-type="fig"}) allowing the creation of many datasets within the same project. Dataset viewing --------------- GeneXplorer is a Perl application that produces a set of html frames that can be used for viewing the expression data (Figure [3](#F3){ref-type="fig"}.). The three frames that it produces are: 1) A radar frame. This frame displays an image map of the data matrix and gives an overall view of the clustered data. The rows correspond to the features or genes (also referred to as reporters), and the columns correspond to the experiments within the dataset. When the image is clicked the next 100 expression patterns starting at the position of the click are displayed in the zoom frame. The position of a bracket on the right side of the radar window indicates the section of the whole radar image that is displayed in the zoom frame. 2) A toolbar frame. Actions in the toolbar may affect either the radar or the zoom frame. There is a tool to set the scaling of the radar image, while the search box allows searching of gene annotations and the expression patterns of the resulting hits are displayed in the zoom frame. In addition the toolbar frame also contains a JavaScript enabled text box that gives feedback depending on the user\'s mouse position, to provide additional information about the genes and experiments within the cluster. 3) A zoom frame. This frame displays a zoomed view of selected expression patterns, such that the user can see both the individual patterns and the associated annotations. The source of the selected patterns can be either a section of the radar image, the result of a search the user performed in the toolbar, or the result of a nearest neighbor search initiated in the zoom frame itself. The expression profiles themselves in the zoom frame are clickable and the resulting search will display the expression pattern for the most similarly expressed genes to the gene that was clicked on, and provide visual feedback as to the level of similarity in their expression profiles. In addition, when the user moves the mouse over parts of the zoom window, additional information is directed back to the textbox in the toolbar. The experiment name, correlation value (Figure [3d](#F3){ref-type="fig"}) and gene annotation is displayed when the mouse is over the experiment image map, the correlation bar, and the expression pattern, respectively. Full text searching ------------------- The search box in the toolbar enables a string search of either all, or specific gene annotation fields. The string may contain more than one term, where each term in the search string should be at least 2 characters long. Spaces between the terms are interpreted as term separators and the terms are combined using the logical \'AND\' operator. Wildcard searches are allowed using the \'\\' character, such that at least one character should precede the wildcard character. The hits resulting from the search are displayed in the zoom frame, as expression patterns. The number of hits displayed in the zoom window is limited to 200 hits. Display configuration --------------------- GeneXplorer allows configurable linking out of the gene annotations to external databases. The number of these links per a gene is not limited, making it easy to be able to look at the information for a gene in several different databases. A configuration file in the dataset directory is used to control where the various gene identifiers are linked. Templates are available for various organisms, and the existing files can be edited manually if a link to a new database is desired. Because of the current limitations of the input cdt file format, setting up the external database links might require manual editing at the time of dataset creation. This is fully described in the README document that is part of the distribution. The external database annotations are not currently updatable in any automated fashion; this will be addressed as part of our plans to make GeneXplorer able to read MAGE-ML (see future plans) that would allow us to do the updates via web services. Installation and use -------------------- The GeneXplorer package is provided as a typical Perl distribution on the Comprehensive Perl Archive Network (CPAN), and adheres to the usual installation mantra of perl modules. After unpacking the software, a user with administrative privileges merely needs to type: perl Makefile.PL make make test make install This will install the libraries and the executable files that are needed for dataset creation by GeneXplorer into the regular system locations, unless otherwise specified during the first step above. The example in Figure [2](#F2){ref-type="fig"} shows the file structure if the library and bin directories under the web server\'s root had been specified for installation of the libraries and executables respectively. To actually use the gx script, it must be copied into a cgi-bin directory, and the various html files must be copied to the appropriate location under the web server\'s root (see Figure [2](#F2){ref-type="fig"}). Results and discussion ====================== In addition to its use within SMD, GeneXplorer has been used by many publications to provide access to microarray datasets through their web supplements, that can be accessed through SMD\'s publication page \[[@B17]\], and was used as the basis for visualization of fuzzy k-means cluster data \[[@B18]\]. We demonstrate on an example dataset how GeneXplorer works \[[@B19],[@B20]\]. Figure [3a](#F3){ref-type="fig"} shows a display of this dataset in the browser window. The whole dataset is displayed in the radar frame, and the zoom window shows the section of this image that was selected, with the gene annotations at a readable size. Clicking on any of the hyperlinks in the zoom frame brings up a new window displaying the biological information for the selected gene that is found in SOURCE (Figure [3b](#F3){ref-type="fig"}.) \[[@B21]\]. Searching the dataset for all the genes whose name field contains the keyword \'kinase\' results in the zoom window shown in Figure [3c](#F3){ref-type="fig"}. This type of search allows comparison of the expression patterns of a subset of the genes based on some functional category -- e.g. GO process-terms, if the annotation fields contain these terms. Clicking on one of the expression profiles (the one belonging to \'Estrogen Receptor 1\', in this case) leads to the display in Figure [3d](#F3){ref-type="fig"}. In the zoom frame it shows the expression profile of the selected gene as the top row, and all the other expression profiles below with Pearson correlation above 0.5. The length of the small orange bar on the right side of the expression profiles gives a graphical representation of these correlation values, while the actual value is displayed in the info box in the toolbar when the mouse is over the orange bar. Future developments ------------------- We are planning to further develop GeneXplorer to enable it to handle other data formats. Specifically, we would like it to be able to accept data files in MAGE-ML format \[[@B22]\], which is becoming a standard file format for communicating gene expression data. In addition, we would like it to be able to display tree views of the clustered data and allow zooming on specific nodes of the cluster. Conclusions =========== We have developed a web-application, GeneXplorer, which allows the visualization of microarray datasets over the Internet using only a web browser. This application has been extremely useful in our experience, where it serves both SMD users during analysis of their data and the public while browsing published datasets. Availability and requirements ============================= GeneXplorer is available at \[[@B23]\] under the MIT Open Source license. It should work on any UNIX-type system capable of running Perl and a Web server, though we ourselves have deployed it on Sun Solaris. Additional information on installation and usage is provided in the installation instructions and documentation that is part of the distribution. List of abbreviations used ========================== SMD: Stanford Microarray Database. Authors\' contributions ======================= CAR designed and wrote the initial version of the GeneXplorer. This was extensively re-factored and modularized by JCM (library modules for dataset) and JD (for explorer). DB was involved in the guidance of the early stages of this project. GS wrote the correlations software and the DataMatrix classes, and guided the development of this project. All authors read and approved the final version of the manuscript. Acknowledgments =============== The authors would like to thank members of the Brown and Botstein laboratories, for their feedback on GeneXplorer, and for providing datasets for testing. Thanks also go to all the members of the Stanford Microarray Database group for stimulating discussions. This work was funded by a grant from the NHGRI, R01HG002732, to GS. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Simplified UML diagram showing the main components of GeneXplorer.The Controller, gx, creates a Microarray::CdtDataset (the Model), which it then uses in construction of a Microarray::GeneXplorer object (the View), which render the Model through a web browser. Microarray::CdtDataset uses PclFile and CdtFile objects, which in turn provide abstraction layers to either a .pcl or a .cdt file, respectively. ::: ![](1471-2105-5-141-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### The file system diagram used by GeneXplorer.Programs within the GeneXplorer distribution assume a certain directory structure during dataset creation, and subsequently during display by the gx application. For dataset creation /html /explorer/ and /data/explorer directories are assumed and dataset specific directories are created under each one of them. For data display web accessible directories are required for reading general images (html/explorer/images/) and images used for a particular dataset (html/explorer/datasets), as well as for creating temporary images (html/tmp). A cgi-bin directory where the gx application itself will exist, and a directory in which the dataset files can be stored and read by the cgi application are also assumed. ::: ![](1471-2105-5-141-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Visualization of a dataset by GeneXplorer.*a.) Overview of GeneXplorer display, explaining the various parts of the window. b.) In the zoom frame, the gene annotations table may contain an unlimited number of graphics and hyperlinks regarding the genes. These are configurable via a server-side stylesheet to accommodate different organism annotation. Clicking on these hyperlinks can take the user to e.g. SOURCE -- on online collection of information about mammalian genes/clones. c.) SEARCH result display in zoom window. The search tool in the tool frame allows searching the annotations by any or all of the fields by using either words or wildcard searches. The result is displayed in the zoom window. d.) GeneXplorer\'s gene correlation display. The genes with expression patterns similar to the gene of interest are presented in an ordered list. The length of the orange bar to the right of the expression data indicates the magnitude of the correlation value. ::: ![](1471-2105-5-141-3) :::	PubMed Central	2024-06-05T03:55:48.613210	2004-10-1	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523853/", "journal": "BMC Bioinformatics. 2004 Oct 1; 5:141", "authors": [ { "first": "Christian A", "last": "Rees" }, { "first": "Janos", "last": "Demeter" }, { "first": "John C", "last": "Matese" }, { "first": "David", "last": "Botstein" }, { "first": "Gavin", "last": "Sherlock" } ] }
PMC523854	Background ========== Complex Regional Pain Syndrome type one (CRPS I) or formerly Reflex Sympathetic Dystrophy (RSD) is a disabling syndrome, in which a painful limb is accompanied by varying symptoms like edema, hyperhidrosis, hypertrichosis, allodynia, coloring of the skin and, over time, atrophy of the involved tissue. Spontaneous recovery does occur and several therapies have been described, but for some patients CRPS I is a chronic disabling disease\[[@B1]\]. Neuropathic pain is a prominent feature of CRPS I, occurring in 75% of cases\[[@B1]\], and many researchers go as far as classifying CRPS I as a neuropathic pain syndrome \[[@B2]-[@B6]\]. Gabapentin (Neurontin^®^, Pfizer) is an anticonvulsant with a proven analgesic effect in various neuropathic pain syndromes \[[@B7]-[@B15]\]. Anecdotal reports suggest that gabapentin may also be an effective analgesic in CRPS patients\[[@B3],[@B8],[@B16]-[@B27]\]. To study this hypothesis, we conducted a double blind, placebo-controlled crossover trial of gabapentin in 58 patients with Complex Regional Pain Syndrome type I. Methods ======= Study population ---------------- This study complied with the Declaration of Helsinki regarding investigations in humans after approval of the protocol by the Institutional Review Board of the University Hospital Maastricht, the Netherlands. Patients were recruited from a database with patients who, in recent years, had been diagnosed with complex regional pain syndrome type I in our hospital. All patients had been treated in our pain management and research center (dept. of Anesthesiology, University Hospital Maastricht, The Netherlands) and had received sympathetic blocks\[[@B28]\], mannitol infusions\[[@B29],[@B30]\], and transcutaneous neuromodulation\[[@B31]\]. All participating patients fulfilled the IASP criteria\[[@B32]\] for the diagnosis of CRPS type I and were included if they were between 18 -- 75 years old and had a score for pain \> 3, as rated on a visual analog score (VAS), where 0 is no pain and 10 is the worst pain imaginable. Apart from IASP criteria, all patients had functional loss and pain outside the original traumatized area. Patients were excluded in case of a possibility of health risk or confounding by other diseases of syndromes, like e.g., pregnancy, known kidney and/or severe liver disease, another (2nd) chronic pain syndrome, known nerve damage in the affected area, active infection or diabetes mellitus. Patients were participating in 8-week periods from 19-11-1998 until 2-12-1999. Gabapentin was not registered as a drug in the Netherlands before or during the trial. After the trial the producing pharmaceutical company supplied gabapentin for compassionate use if indicated. Treatment --------- Since our patient population consisted of chronic CRPS I patients with a multiple years\' duration of pain complaints refractory to various treatments, we assumed that their pain complaints would be more or less stable. We therefore undertook a double blind, randomized crossover study. Randomization of patients took place after baseline measurements and written informed consent. The assignment scheme was generated by our hospital pharmacy from a table of random numbers. The closed envelopes containing the assignments were prenumbered and kept at the pharmacy. The first treatment group received gabapentin, followed by a washout period and placebo treatment. The second treatment group received placebo treatment, which was followed by a washout period and gabapentin treatment. Each medication period lasted three weeks separated by the two-weeks washout period. Medication was stocked and delivered to the patient at the hospital pharmacy. Both the gabapentin capsules and the identical placebo capsules were delivered immediately before the start of the two medication periods. Left over medication was recollected and counted. The gabapentin (GBP) dose was slowly increased to reduce adverse side effects: 600 mg\'s GBP AN once a day on day 1--2 600 mg\'s GBP b.i.d. on day 3--4 600 mg\'s GBP t.i.d on day 5--21 Placebo dose was identically titrated. Patients were allowed to take their usual analgesics and were told preferably not to change the usual dose. Follow-up measurements ---------------------- The patients were reevaluated at the hospital three weeks (T1), five weeks (T2) and eight weeks (T3) after randomization. During the trial, the patient noted her/his pain rate of the past 24 hr (VAS) and the use of additional analgesics in a diary. During each hospital visit the following assessments were done: 1\. Global perceived effect (GPE) on pain indicating: worst ever; much worse; somewhat worse; not improved/not worse; somewhat improved; much improved and best ever. GPE on function was scored on an analogous scale. 2\. Neuropathic pain scale (NPS), a 10 item qualitative evaluation of neuropathic pain\[[@B33]\]. 3\. Sensibility through Von Frey monofilament skin application each on 9 areas corresponding to cutaneous nerve branches and dermatomes of either both hands or both feet\[[@B34]\]. Stimulus placement of filaments was as follows: one second for placement, one second for bending and one second for removal. (handset with resp. 0.0677, 0.4082, 2.052 and 3.632 grams calculated force, North Coast Medical, Inc., San José, USA). 4\. Mechanic allodynia test with brush strokes and static pressure with the finger tip\[[@B35]\], on 9 areas corresponding to cutaneous nerve branches and dermatomes of either both hands or both feet. 5\. Edema, discoloration, and changed skin temperature were scored after physical examination on a three point scale indicating no, some or overt presence of each sign, the latter two signs in comparison to the healthy or healthiest limb. Physical examination in CRPS is well comparable to instrumental evaluation of signs with volumeter, infrared thermometer and goniometer\[[@B36]\]. 6\. Impairment and disability tests: Symptom Checklist-90-Revised (SCL-90-R)\[[@B37]\], Brief Pain Inventory\[[@B38]\], adapted for CRPS to measure the influence of CRPS in general on daily life by 0--10 scale ranging from 0 (\'CRPS has not interfered\') to 10 (\'CRPS completely interfered\'), \'range of motion\' as a parameter of limb function. Side effects during treatment ----------------------------- A blinded independent investigator (STvdB) did sensibility, allodynia and range of motion tests (see above). A physician (AvdV), who examined each patient, did all the other measurements throughout the trial. Patient, investigator and physician were unaware of the treatment received. We tested blinding by questioning physician and participants after each medication period. Statistical analysis -------------------- The statistical analysis of VAS-scores was determined per patient using estimating medication and period effect through linear regression analyses. Possible relationship of patient characteristics and outcome was tested by Pearson R\'s test. Mann-Whitney analyses were used for monofilament sensitivity on log-transformed data. Three point scales and seven point scales were dichotomized and like the SCL-90-R, NPS and CRPS-Brief Inventory questionnaires intra-individual paired tested (McNemar, t-test, Bonferroni-Holm corrected for multiple tests). Student t-tests and regression analysis were used to test treatment effect, which is calculated in crossover studies as ((AT1-AT0)-(AT3-AT2))/2+((BT3-BT2))-BT1-BT0)/2, where A represents data of placebo starters and B data of gabapentin starters both before (T0,T2) and after (T1,T3) treatments\[[@B39]\]. Blinding was tested with Chi-square analyses. Possible related factors to therapy effect were analyzed with forward stepwise logistic regression. Data analyses required complete data sets. Patients who were not completing one or two treatments were excluded for analyses. We tested two-tailed, with α = 0.05 as a level of significance (Excel 2000, SPSS 10.0 for Windows). Results ======= Demographics ------------ After randomization 58 patients were enrolled, with a mean age of 44.0 (range 24--75) resulting in 29 patients in the gabapentin-placebo arm and 29 patients in the placebo-gabapentin arm; 49 patients completed the gabapentin period, 50 patients the placebo period, 46 patients completed both periods and were used for further within-patient paired analysis (Fig. [1](#F1){ref-type="fig"}). Twelve patients discontinued treatment of which 6 during the placebo treatment, 2 during washout and 4 during GBP treatment. Three of these four GBP users discontinued due to side effects (Fig. [1](#F1){ref-type="fig"}). Between randomization and start of (placebo) medication one patient withdrew after rereading the information letter about possible side effects. These patients were excluded from analysis, since intra-individual testing was necessary for most of the data-analyses. Patients, who could not be used for analysis, did not differ in their characteristics from the total group nor comparing between the two arms of treatment (Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). When comparing the placebo-GBP arm and GBP-placebo arm on sexes, age and pain level before period 1 or 2, duration of illness, SCL-90-R score, we could not find a difference between the two arms (Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). SCL-90-R score revealed increased values on any subscale comparing to standard norms, indicating personal distress (Table [2](#T2){ref-type="table"}). We found relative higher scores on somatic and sleeping complaints. The SCL-90-R scores were identical to control chronic pain patients (N = 143), besides higher score on sleeping complaints (T. Forouzanfar, data not published). Trial medication was returned and counted afterwards, but revealed no lack of compliance in any patient. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Selection of patients participating in the trial ::: ![](1471-2377-4-13-1) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Patient characteristics ::: Placebo starter GBP starter ------------------------- --------------------- ----------------- ------------------------ ----------------- Excluded from analysis Placebo starter GBP starter N = 24 N = 22 N = 5 N = 7 Sex (F/M) 21/4 18/4 3/2 6/1 Age in years 42 (± 13) 47 (± 14) 40 (± 11) 43 (± 11) Duration in months 43 (± 36) 44 (± 21) 83 (± 39) 45 (± 30) VAS0 64.2 (± 16) 62.5 (± 18) 62 (± 10) 67 (± 12) VAS2 67 (± 20) 64(± 21) Upper extremity in pain 3R 8L 3RL = 14 8R 7L 4RL = 18 2R 0L 1RL 2R 4L 0RL Lower extremity in pain 2R 7L 4RL = 13 3R 3L 0RL = 6 2R 1L 0RL 1R 1L 1RL R/L/RL represents no. of patients that report pain in resp. right, left or bilateral extremities. A few patients had upper and lower extremity pain. VAS is pain level on visual analogue scale. VAS0 is day 1, VAS2 is day 21 (post wash-out). Data are mean with (SD) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Basic characteristics of participating patients on neuropathic pain scale (NPS), CRPS brief inventory and SCL-90-R. Data are mean with (± standard deviation). ::: NPS Intensity Sharpn. Hot Aching Cold Sens. Itch. Comfo. Deep p. Superf.P. ------------------- ----------------- ----------------- ----------------- ----------------- ----------------- ----------------- ----------------- ----------------- ----------------- ------------------ Mean N = 24 7,3 (± 1,8) 7,3 (± 1,5) 6,0 (± 3,2) 7,0 (± 2,4) 6,0 (± 3,2) 6,6 (± 2,5) 2,8 (± 2,5) 7,8 (± 1,6) 7,6 (± 1,3) 6,0 (± 2,7) Mean N = 22 7,3 (± 1,4) 7,4 (± 1,5) 5,9 (± 3,1) 7,2 (± 1,6) 5,9 (± 3,1) 7,2 (± 2,5) 3,7 (± 3,1) 7,7 (± 1,4) 7,8 (± 1,2) 6,8 (± 2,4) Lost ABN = 5) 7,6 (± 0,5) 8 (± 1,2) 6,2 (± 3,8) 6,8 (± 1,8) 8,2 (± 0,8) 8,4 (± 0,9) 2,8 (± 3,8) 7,3 (± 1,7) 8,2 (± 1,3) 7,4 (± 1,1) Lost N = 7 7,7 (± 1,4) 7 (± 1,3) 5,7 (± 3,3) 8,3 (± 1,0) 8,4 (± 1,7) 8,6 (± 1,1) 2,3 (± 2,9) 8,8 (± 1,0) 8,3 (± 1,1) 8,6 (± 0,8) SCL-90-R anxiety fobic depression somatiz Obs-comp sensitivity hostility insomnia psneu Total AB (n = 24) 15,9 (± 5,8) 10,9 (± 4,0) 31,1 (± 11,4) 26,2 (± 8,3) 19,2 (± 7) 26,8 (± 10,1) 10,2 (± 5,4) 10,1 (± 3,9) 163,0 (± 47,5) Total BA (n = 21) 15,9 (± 5,8) 11,2 (± 6,3) 33,6 (± 14,4) 25 (± 8) 19,7 (± 6,0) 28,1 (± 10,5) 8 (± 1,9) 10,3 (± 3,5) 163,8 (± 44,9) Lost AB (n = 5) 18,4 (± 11,2) 13,4 (± 8,3) 27,8 (± 10) 26,8 (± 7,8) 21 (± 2,9) 26 (± 2) 9,2 (± 2,8) 11,6 (± 2,1) 166,8 (± 41,2) Lost BA (n = 6) 19,7 (± 11,1) 12,2 (± 4,4) 33 (± 17,5) 29,17 (± 11,5) 22,3 (± 9,8) 33,5 (± 18,0) 11,7 (± 7,5) 12,8 (± 2,1) 189,7 (± 80,1) CRPS brief inventory= = = = = = = = = = [1]{.underline} [2]{.underline} [3]{.underline} [4]{.underline} [5]{.underline} [6]{.underline} [7]{.underline} [8]{.underline} [9]{.underline} [10]{.underline} Mean N = 24 7,4 (± 1,7) 6,5 (± 2,1) 6,3 (± 3,1) 7,6 (± 2,0) 4,2 (± 2,6) 7,0 (± 2,8) 6,4 (± 2,2) 4,3 (± 3,4) 7,2 (± 2,2) 5,6 (± 2,5) Mean N = 22 7,0 (± 2,0) 5,0 (± 3,0) 6,3 (± 3,0) 7,8 (± 2,3) 4,8 (± 2,9) 7,7 (± 2,1) 5,1 (± 3,1) 5,1 (± 2,7) 6,8 (± 2,5) 5,5 (± 2,5) Lost AB n = 5 7,4 (± 1,8) 7,2 (± 1,6) 7,8 (± 0,4) 9 (± 1) 4,6 (± 1,5) 8,2 (± 0,8) 6 (± 2,6) 6,4 (± 2,6) 7,2 (± 3,6) 6,8 (± 1,1) Lost BA n = 7 8,6 (± 1,4) 6 (± 3) 6,4 (± 3,9) 9 (± 1,2) 5 (± 2,9) 8 (± 1,6) 7 (± 2,8) 6,7 (± 1,8) 7,3 (± 2,3) 6,6 (± 2,4) *NPS*description of pain in terms of 1. intensity 2. sharpness 3. hot 4. aching 5. cold 6. sensitive 7. itching 9. comfortability 10a. intensity deep pain 10b. intensity superficial pain. Item 8 is a nominal scale left out of analysis. *Symptom Checklist-90-Revised (SCL-90-R)*subscale on anxiety, phobic anxiety, depression, somatization, obsessive compulsive, interpersonal sensitivity, hostility, insomnia and psycho neuroticism. *CRPS BI: influence of CRPS on 1. general activity 2. mood 3. mobility 4. normal work 5. personal relationships 6. sleep 7. enjoyment of life 8. self care 9. recreational hobbies 10. social activities. CRPS BI and NPS on a 0--10 scale* ::: Blinding -------- After each medication period both patient and physician were asked about their ideas concerning study medication in the past period. The treating physician guessed the used medication correctly more often after both phases than can be explained by coincidence (p = 0.000). Blinding for patients was sufficient in the first phase, but not anymore after the second phase (p = 0.2 versus p = 0.000). Response to treatment --------------------- ### Pain Comparing gabapentin and placebo users in terms of pain relief, there was a significant pain relief in favor of gabapentin in the first period. Therapy effect in the second period was less, finally resulting in no significant effect combining results of both periods. There was an unexpected increase of pain level above baseline in the washout period for both the gabapentin starters and placebo starters (Figure [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### VAS for pain in both groups at start (T0), three weeks (T1), five weeks (T2), and eight weeks (T3) after randomization. T0-1 is the first treatment period, and T2-3 the second ::: ![](1471-2377-4-13-2) ::: Global perceived pain relief as measured by the seven-point scale showed a significant effect for gabapentin, and also more pronounced in the first period. This measurement also found a significant effect in the second period, with an effect being defined as a patient scoring \'much improvement\'. Statistical analysis of global perceived effect showed significant more treatment effect (p = 0.002) with 43 % versus 17 % reported pain relief respectively during gabapentin compared to placebo treatment. 13 % of patients reported aggravation of pain during gabapentin vs. 9 % during placebo treatment (Figure [3](#F3){ref-type="fig"} and table [3](#T3){ref-type="table"}). Stepwise forward logistic regression analysis of baseline value of pain level, age, sex, duration of illness, location of illness, mono- or bilateral CRPS, trial arm and all items of CRPS-BI, NPS and SCL-90-R was performed. Only the level of self care was related to perceived pain relief during GBP. The neuropathic pain scale, indicating different aspect of pain, improved significantly in terms of less hot and more comfortable, but not when corrected for multiple tests (Bonferroni-Holm correction). We found that during gabapentin use, patients reported equal use of co-medication comparing to baseline assessment and placebo-use with a non-significant trend towards less medication during GBP use. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Global perceived pain relief (on a seven-point scale) as reported by patients. GBP-1 and -2 denote patients receiving GBP in the first and second period; placebo-1 and -2 are analogously denoted. ::: ![](1471-2377-4-13-3) ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Patients (%) with global perceived effect on pain in the four arms of treatment and totals for the two treatments. ::: Treatment period GBP-1= Placebo-1= Wash-out= GBP-2= Placebo-2= ------------------------- ---------------- --------------- ------------------- ------------ ------------ \% some improvement (n) 45% (10) 13% (3) 1 8% (2) 13% (3) \% much improvement (n) 14% (3) 5% (1) 0 21% (5) 4% (1) \% total (n/N) 59% (13/22) ^a^17% (4/24) 1 29% (7/24) 18% (4/22) = Total = GBP = = Total = placebo = = = \% some improvement 26% (12/46) 13% (6/46) \% much improvement 17% (8/46)β 4% (2/46) \% total (n/N) 43% (20/46)^a^ 17% (8/46) worsened 13% (6/46) 9% (4/46) GBP-1 is gabapentin treatment before wash-out. GBP-2 is gabapentin treatment after wash-out. \'α\' is significant, P \< 0.005, \'β\' is P \< 0.10 McNemar two sided tested gabapentin versus placebo. ::: Sensory tests ------------- Each participant was tested throughout the study on response to mechanical stimuli with von Frey filaments. The CRPS patients had sensory deficits at baseline. Application of smaller filaments was not felt in multiple skin areas. We found, with Mann-Whitney analyses, that this sensory deficit was significantly reversed in gabapentin users in comparison to placebo users (p = 0.027). This difference was found in patients with upper and lower extremity CRPS, but was still significant in the subgroup of lower extremity CRPS (p = 0.011) as seen in table [4](#T4){ref-type="table"}. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Mann-Whitney scores of monofilament application in CRPS patients testing cutaneous sensibility thresholds ::: Mean ranking ---------------- ------------------ --------------- ---------------- Hand Feet Total Placebo 12.0 (N = 12) 5.5 (N = 10) 16.8 (N = 22) Gabapentin 15.6 (N = 15) 12.0\(N = 3) 25.0\(N = 18) Significant different values (p \< 0.05) are marked with\. ::: Mechanical allodynia to static and dynamic stimuli (soft touch and brush) was measured by a mean of 11-point scales (range 0--10) of 9 areas of the hand/feet corresponding to cutaneous nerve branches. We found no effect of gabapentin on allodynia in comparison to placebo. Other symptoms -------------- No difference was found on the parameters edema, discoloration, range of motion of wrist/ankle and fingers/toes between placebo and GBP. 10 patients out of 45 improved in relative skin temperature during placebo use compared to 18 patients out of 45 in gabapentin, which is two sided tested not significantly different (McNemar analysis, p = 0.096). Limb dysfunction and quality of life ------------------------------------ The reported function improvement was, with 10 positive responders during GBP versus 7 positive responders during placebo, not significantly different (N = 46) between the two treatments. The SCL-90 showed no significantly better scores during gabapentin treatment. CRPS-BI showed improvement of sleep between placebo treatment and gabapentin treatment., but this effect disappeared after Bonferroni-Holm correction. Adverse effects --------------- Dizziness, somnolence and lethargy were significantly more often reported during gabapentin use than during placebo. Before washout 95 % of patients (n = 21) reported side effects during gabapentin use versus 58 % in placebo treatment (n = 14). After washout this was respectively 63% (n = 15) in GBP and 32% (n = 7) in placebo use. For more details on side effects see table 6. Since a high incidence of side effects can produce a stronger placebo effect, we analyzed the possible correlation between side effects and pain relief. There was a small relation between perceived side effects and pain relief in placebo users in period 2 (p = 0.04, Pearson\'s R value is 0.4), but, whether in period 1 or period 2, no relationship was found during the use of gabapentin (p = 0.2 in period 1, P = 0.4 in period 2). Discussion ========== To evaluate gabapentin treatment as a treatment for pain in CRPS, we conducted a placebo-controlled crossover study. We conclude from our trial that overall, gabapentin did not relieve pain as compared to placebo on pain visual analogue scores, our primary outcome measure. Gabapentin relieved pain in a subgroup of patients and gave a significant global perceived pain relief. The effect was mild and there was no patient in which gabapentin completely eliminated pain. Moreover, the frequency of side effects as dizziness, somnolence and lethargy was higher during gabapentin treatment than with placebo. These side effects probably also account for the relative lack of blinding we observed in our study. This does not mean that the study was biased: our population was chronic CRPS patients who all had undergone numerous unsuccessful treatments, and clearly wanted the drug to work. Any possible bias would therefore have been positive towards gabapentin. Although we did not find a significant pain reducing effect when analyzing the complete trial, we did find a significant effect in the first half of the trial. In fact, the difference in outcome between the two trial halves is striking. There was a reverse carry-over effect resulting in increasing pain above baseline after the washout period for both gabapentin and placebo starters. The increase of pain intensity above baseline level in the second period (before the start of placebo treatment) cannot be explained pharmacologically. Gabapentin has no known biological dependency or tolerance mechanism. It can be a period effect, although this would more likely result in a regression to the middle instead of increasing pain. Perhaps this is a reversed placebo effect in which the expectation and/or the actual perception of not receiving gabapentin anymore might increase pain intensity. Kemler and de Vet found that treatment allocation in a trial could influence pain intensity in CRPS\[[@B40]\]. The decreasing therapy effect after washout is found in other crossover pain trials\[[@B41]\]. Expectation and attention have been shown to be powerful influences on pain pathways in the brain\[[@B42]\], and perhaps a crossover design is not suited to study treatments in chronic pain patients. We found a decreased sensory deficit in gabapentin users compared to placebo users. We did not expect this, but found in the literature several cases in which gabapentin decreased the area of hypesthesia in neuropathic pain syndromes\[[@B43]\]. This has, to our knowledge, never been described for any other medication. Numbness or mechanical hypesthesia is a frequently found complaint for approximately 75 % of CRPS patients, which can improve in time spontaneously and after placebo treatment\[[@B44]\]. It is possible that the somatosensory findings and pain outside the original area of trauma can be attributed to referred pain mechanisms. Gabapentin has been reported to alleviate referred pain\[[@B45]\]. Since many CRPS patients have mechanical hypesthesia, we hypothesize that gabapentin influences some common neural pathway for \'referred\' sensations, whether mechanical sensation or pain. Conclusions =========== Gabapentin had a mild effect on pain in patients CRPS I. It significantly reduced the sensory deficit in the affected limb. A subpopulation of CRPS patients may benefit from gabapentin, but then for each individual patient the benefit has to be weighed against the frequently occurring side effects. Competing interests =================== Parke-Davis (now a Warner-Lambert/Pfizer division) supplied gabapentin and matching placebo capsules for this trial. Drs. Van de Vusse and Weber have received financial support from Parke-Davis to attend one congress. Parke-Davis has had no role in the writing of this manuscript Authors\' contributions ======================= AvdV initiated the trial and wrote, with WEJW and AHFK, the protocol. The study and its data management was done by AvdV and SS-vdB. AHFK did the statistical analyses. AvdV wrote the first draft of the manuscript, which was finished in its final form by WEJW. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Side effects as mentioned after treatment ::: Adverse effect* Gabapentin (N = 54) n (%) Placebo (N = 51) n(%) Significance -------------------- ------------------------------- --------------------------- ------------------ Dizziness 20 (37.3) 2 (3.9) P = 0.0000 Somnolence 15 (27.8) 3 (5.9) P = 0.003 Lethargy 11 (20.4) 1 (2.0) P = 0.003 Nausea 10 (18.5) 5 (9.8) n.s. Headache 8 (14.8) 3 (5.9) n.s. Stomach problems 4 (7.4) 3 (5.9) n.s. \'drunken\' 4 (7.4) 0 (0) n.s. Disturbed gait 4 (7.4) 0 (0) n.s. Water retention 1 (1.9) 3 (5.9) n.s. Data on all patients who started treatment and returned for assessment after 3 weeks, with or without completing 3 weeks of treatment. n.s. is \'not significant\' ::: Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2377/4/13/prepub> Acknowledgements ================ We would like to thank Inge Lamé for her support in data management, Drs. Brad Galer and Mark Jensen for the NPS and CRPS-Brief Pain Inventory.	PubMed Central	2024-06-05T03:55:48.615028	2004-9-29	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523854/", "journal": "BMC Neurol. 2004 Sep 29; 4:13", "authors": [ { "first": "Anton C", "last": "van de Vusse" }, { "first": "Suzanne GM", "last": "Stomp-van den Berg" }, { "first": "Alfons HF", "last": "Kessels" }, { "first": "Wim EJ", "last": "Weber" } ] }
PMC523855	Background ========== An extremely broad variety of phenomena in physics, biology, and the social sphere is described by power law distributions. The power laws apply to the distribution of the number of links between documents in the Internet, the population of towns, the number of species that become extinct within a year, the number of sexual and other contacts between people, and numerous other quantities \[[@B1]-[@B4]\]. In the field of genomics, the \"dominance by a selected few\" \[[@B5]\] encapsulated in the power laws applies to the distribution of the number of transcripts per gene, the number of interactions per protein, the number of genes in coexpressed gene sets, the number of genes or pseudogenes in paralogous families, the number of connections per node in metabolic networks, and other quantities that can be obtained by genome analysis \[[@B5]-[@B9]\]. Mathematically, these distributions are described by the formula: P(i) ≈ ci^-γ^where P(i) is the frequency of nodes with exactly iconnections or sets with exactly imembers, γis a parameter which typically assumes values between 1 and 3, and cis a normalization constant. Obviously, in double-logarithmic coordinates, the plot of Pas a function of iis close to a straight line with a negative slope. Recently, it has been shown that the distributions of several genome-related quantities are best described by the so-called generalized Pareto function: P(i) = c(i+ a)^-γ^where γ\> 0, aare parameters \[[@B10]-[@B13]\]. At large i(i\>\>a), this distribution is indistinguishable from a power law, but at small i, it deviates substantially, with the magnitude of the deviation depending on a. Power law distributions and the associated scale-free networks are compatible with the intuitively plausible mechanism of evolution by preferential attachment although other modes of evolution are also possible \[[@B9],[@B14]\]. Under preferential attachment, a network or a mathematically analogous object, such as an ensemble of gene families, grows via attachment of new nodes to the pre-existing ones with a probability that is proportional to the degree (number of connections) of the latter. However, preferential attachment or other general evolutionary principles associated with power law type distributions and scale-free phenomena do not actually explain the emergence of these phenomena in biologically meaningful terms. A biological explanation involves, at a minimum, identifying the elementary events underlying the evolutionary process and the simplest models of evolution that include these events and are compatible with the observations. Under this logic, families of paralogous genes represent a perfect object for evolutionary modeling. Indeed, for these families, elementary evolutionary processes are defined naturally. By definition, paralogous families evolve by gene duplication. It has been long suspected and, with the advent of genomics, established beyond reasonable doubt that genome evolution proceeds largely by duplication of genes or portions thereof, and even long genomic segments or entire genomes \[[@B15]-[@B20]\]. All sequenced genomes contain numerous paralogous genes, and in more complex genomes, the majority of genes have at least one paralog \[[@B21],[@B22]\]. Duplication is followed by mutational diversification and gradually leads to functional differentiation of the paralogs. It is thought that such differentiation occurs via the routes of neofunctionalization (emergence, in one of the paralogs, of a new function non-existent in the ancestral gene) \[[@B16]\] and, probably most often, subfunctionalization, i.e., partitioning of subfunctions of the ancestral gene among the paralogs \[[@B23],[@B24]\]. Hence, duplicationobviously is the first elementary process of genome evolution. Genomes and gene families not only grow but often shrink or, probably most of the time, persist in equilibrium. Therefore, duplication must be counter-balanced by the opposite elementary process, gene loss. Again, comparative genomics has shown that gene loss occurs in all species and seems to be extensive in certain lineages, particularly in parasites \[[@B25]-[@B27]\]. Finally, genes new to a given lineage may emerge either as a result of a dramatic change after duplication obliterating all \"memories\" of a gene\'s origin, or via horizontal gene transfer, or by evolution of a protein-coding gene from a non-coding sequence (rare as this latter process might be). Collectively, the contribution of these processes to genome evolution may be termed innovation. Gene duplication, gene loss, and innovation seem to comprise a reasonable minimal set of elementary events for modeling genome evolution. The only potential major addition could be rearrangement of the gene structure whereby genes accrete or lose domains. However, at least for first approximation modeling, these changes could be covered either by duplication, if they do not yield new genes without detectable relationships to pre-existing families, or by innovation if they do. We should further note that evolutionary analysis of paralogous gene families can be reasonably viewed as a study of the evolution of genomes themselves if all genes are viewed as members of paralogous families, ranging in size (number of members) from 1 to N(the size of the largest family). Of course, one must keep in mind that describing genome evolution in terms of gene duplication, loss, and innovation represents a high level of abstraction, whereby a gene is considered an atomic unit of evolution, and mutation processes occurring within a gene are ignored. However, numerous comparative-genomic studies have shown the utility of the gene-level abstraction both for systematic prediction of the functions of uncharacterized genes using the patterns of their distribution in diverse genomes \[[@B28]-[@B31]\] and for understanding general evolutionary trends. A striking recent example of the latter type of achievement is the demonstration that different functional categories of genes scale differently with genome size, with the steepest ascent of regulatory genes offering a plausible explanation for the observed limits of genome size in prokaryotes \[[@B32]\]. A natural framework for modeling evolution of gene families is a birth-and-death process, a concept well explored in many physical and chemical contexts \[[@B33]\]. Duplication constitutes a gene birth, and gene loss is a death event; innovation also can be readily incorporated in this context. The birth-and-death approach has been applied to modeling the evolution of paralogous genome family sizes \[[@B6],[@B12],[@B34]\], the distribution of folds and families in the entire protein universe \[[@B35]\], and protein-protein interaction networks \[[@B36],[@B37]\]. For over a century since the publication of Darwin\'s seminal work \[[@B38]\], biologists believed that evolution at all levels is largely driven by natural selection \[[@B39]\]. However, the advent of molecular evolution shifted the perspective by demonstrating, largely through the work of Kimura and his school, that many, if not most, of the fixed nucleotide substitutions are effectively neutral \[[@B40]\]. Recent comparative analyses of gene expression led to the expansion of the neutral evolution concept beyond the genome sequence, at least to the level of the transcriptome \[[@B41],[@B42]\]. Perhaps the principal importance of the neutral theory is that it leads to a change of the prevailing null hypothesis of evolutionary biology: neutrality should be taken as the null hypothesis, and selection should be invoked only when this hypothesis can be rejected. Birth and death models naturally fit this paradigm because they do not include the notion of selection (at least not explicitly). It is therefore of considerable interest to determine whether or not simple models of this class can be rejected as the explanation for various observed features of genomes. In the previous work \[[@B12]\], we examined in detail simple deterministicmodels of genome evolution, which we dubbed BDIMs, after [b]{.underline}irth (duplication), [d]{.underline}eath (elimination), and [i]{.underline}nnovation (de novoemergence or acquisition via horizontal gene transfer) models. We showed that the power law asymptotic of the size distribution of gene families appears if, and only if, birth and death rates of domains in families of sufficiently large size are balanced (asymptotically equal up to the second order) and that any power asymptotic with γ≠ 1 appears only if the per gene birth/death rates depend on the size of the gene family. We showed that the simplest model that adequately approximates the empirical data on gene (domain) family size distributions is the linear2^nd^order balanced BDIM. Subsequently, we expanded the BDIM framework by introducing stochastic BDIMs, which account not only for the stationary state of the gene ensemble but also for the characteristics of evolution of the system, such as the probability of the formation of a family of the given size before extinction and the mean times of formation and extinction of a family of a given size \[[@B43]\]. We first investigated these issues for the linear 2^nd^order balanced stochastic BDIM. Given the published estimates of the rates of gene duplication and loss \[[@B24]\], we found that this version of BDIM, which gives a good approximation of the stationary distributions of family sizes for different genomes, predicts completely unrealistic mean times for reaching the observed sizes of the largest domain families. In computer simulations with a large ensemble of genes, even the minimum time required for the formation of the largest family was shown to be unrealistically long. Thus, the linear BDIM is incompatible with the estimates of the rate of genome size growth derived from the empirical data. Therefore we performed a preliminary examination of non-linear, higher degree BDIMs and showed that the rate of genome size growth increases with the degree of the model, rendering non-linear BDIMs more realistic models of genome evolution \[[@B43]\]. Here, we present a detailed analysis of the properties of different non-linear stochastic BDIMs, including polynomial, rational, and logistic ones, which were obtained by the appropriate transformations of the original linear model. These models generated the same stationary family size distribution, but the stochastic properties of the higher order models were dramatically different from those of the linear BDIM. The mean number of elementary events, duplications and deletions, which are required for the formation of the largest family, decrease monotonically with the increase of the model degree. By contrast, the mean time of formation of a gene family of the given size under a fixed average duplication rate went through a minimum depending on the model degree; typically, the model degree corresponding to this minimum was between 2 and 3. However, even with this optimal degree, the mean times of formation of the largest families in different genomes were unrealistically long. The times of formation and extinction of gene families are random variables with unknown distributions. Therefore it was important to determine the variance of these times and the number of elementary events preceding the formation and extinction of the largest families. We found that the coefficients of variation were very large such that the extreme values of the formation times for the largest family could differ from the mean time by at least two orders of magnitude. Thus, for assessing the feasibility of the formation of the largest families under a given model, the relevant value is not the mean but the minimal time of family formation over the entire ensemble of genes. Using Monte Carlo simulations, we show that the minimal time required for the formation of families of the expected size under BDIMs of the orders between 2--3 is compatible with the timescale of genome evolution. Results and discussion ====================== 1. Definitions and empirical data --------------------------------- ### The basic BDIM definitions and assumptions We treat a genome as a \"bag\" of genes (or, more precisely, portions of genes) encoding protein domains (or simply domainsfor brevity; see \[[@B12]\] for details). Domains are treated as independent evolving units disregarding co-occurrence of domains in multidomain proteins. Each domain is considered to be a member of a family, which may have one or more members. In this work, we interchangeably refer to domain families or gene families. Three types of elementary events are postulated: i) birth, which yields a new member in the same domain family as a result of gene duplication, ii) death, i.e., inactivation and/or deletion of a domain, and iii) innovation, which generates a new, single-member family. Innovation may occur via domain evolution from a non-coding sequence or a non-globular protein sequence, via horizontal gene transfer from another species, or via radical change of a domain after duplication. The rates of elementary events are defined as the probabilities of the respective events during an infinitesimally short time interval \[[@B44]\] and is postulated to be independent of time (all analyzed models are homogeneous) and of the structure, biological function, and other features of individual domain families. Clearly, these assumptions are simplifications made in order to avoid prohibitively complex models; the justification is that, over large (genome-wide) ensembles of families and long time intervals, the existing non-homogeneities are likely to cancel out, making homogeneous models realistic. It may be useful to emphasize that homogeneity of the models does not imply constancy of the number of events during any finite time interval, which is a random variable. The data on the size of domain families in sequenced genomes were obtained as described previously \[[@B12]\]. Briefly, the domains were identified by comparing the CDD library of position-specific scoring matrices (PSSMs) for domains extracted from the Pfam and SMART databases, to the protein sequences from completely sequenced eukaryotic and prokaryotic genomes [http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db =Genome](http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db =Genome) using the RPS-BLAST program \[[@B45]\]. In a finite genome, the maximum number of domains in a family cannot exceed the total number of domains and, in reality, is probably much smaller. Let Nbe the maximum possible number of domain family members (this limit is introduced for technical reasons; however, this should not be perceived as a biologically unrealistic assumption because Ncan be made extremely large, e.g., to exceed the number of genes in the largest known genome by several orders of magnitude; furthermore, almost all of the results below are valid with N= ∞ under certain well defined conditions, which ensure the existence of the ergodic distribution of the birth-and-death process). We also consider virtual, \"empty\" families that consist of 0 domains. In the stochastic BDIMs, newborn domains are extracted from this class and dead domains return to it. Originally, we examined exclusively the deterministic version of the BDIMs \[[@B12]\]. Introduction of the 0 class \"closes\" the model and allows us to transform it into a Markov process, which provides for the possibility to explore the stochastic properties of the system \[[@B43]\]. In these stochastic models, innovation was not introduced explicitly as it was in the deterministic models, but was implied in the emergence of domains from the 0 class. Let p~i~(t) be the frequency of a domain family of size i. Then p~i~(t) satisfy a system of forward Kolmogorov equations for birth-and-death process (e.g., \[[@B44],[@B46]\]): dp~0~(t)/dt= -λ~0~p~0~(t) + δ~1~p~1~(t), dp~i~(t)/dt= λ~i-1~p~i-1~(t) - (λ~i~+ δ~i~)p~i~(t) + δ~i+1~p~i+1~(t) for 0 \<i\<N, (1.1) dp~N~(t)/dt= λ~N-1~p~N-1~(t) - δ~N~p~N~(t). Mathematically, (1.1) defines the state probabilities of a birth-and-death process with the finite number of states {0,1,\...N} and reflecting boundaries in 0 and N. The evolution of individual trajectories of the birth-and-death process X(t), whose state probabilities satisfy the system (1.1), can be described as follows. At the starting time, the system is situated in some initial state x~0~. The time axis {t≥ 0} can be divided into intervals \[0,τ~1~), \[τ~1~, τ~2~), \[τ~2~, τ~3~) \... such that X(t) is a constant on each interval. If, at the moment τ~n~, the system was situated in the point x~n~= i, then, in the moment τ~n+1~, the system moves either into the state i+1 with the probability β~i~= λ~i~/(λ~i~+ δ~i~) or into the state i-1 with the probability μ~i~= δ~i~/(λ~i~+ δ~i~). The sojourn time t~i~= τ~n+1~- τ~n~between the arrival at the point x~n~= iand the exit from this point is a random variable independent of the previous history of the system and is distributed exponentially: P{t~i~≥ x} = exp(-(λ~i~+ δ~i~)x). Note that the random variables t~i~are independent, and the mean sojourn time, E(t~i~), in the state iis E(t~i~) = 1/(λ~i~+ δ~i~). Process (1.1) has a unique stationary ergodic distribution p~0~,\...,p~N~defined by the equalities dp~i~(t)/dt= 0 for 0 ≤ i≤ N. Let J(i, t) = δ~i~p~i~(t) - λ~i-1~p~i-1~(t) be the currentthrough the state iin ttime moment, J(i) = δ~i~p~i~- λ~i-1~p~i-1~be the current in the stationary state. Then the equation for the stationary distribution can be written as J(i+1) - J(i) = 0. As the system is closed, J(0) = 0 and hence J(i) = 0 for all i, such that p~i~/ p~i-1~= λ~i-1~/δ~i~. ![](1471-2148-4-32-i1.gif) ![](1471-2148-4-32-i2.gif) We will consider also the variant of this model with states {1,\...N} and reflecting boundaries in states 1 and N: dp~1~(t)/dt= -λ~1~p~1~(t) + δ~2~p~2~(t), dp~i~(t)/dt= λ~i-1~p~i-1~(t) - (λ~i~+ δ~i~)p~i~(t) + δ~i+1~p~i+1~(t) for 1 \<i\<N, (1.3) dp~N~(t)/dt= λ~N-1~p~N-1~(t) - δ~N~p~N~(t). This model describes the evolution of the size of a domain family that includes an indispensable (essential) gene and is not allowed to go extinct. Similarly, for model (1.3), the ergodic distribution is: ![](1471-2148-4-32-i3.gif) ![](1471-2148-4-32-i4.gif) The ergodic distribution (1.2) (or 1.4) is globally stable and is approached exponentially with respect to time from any initial state. The asymptotic of the ergodic distribution is completely defined by the asymptotic behavior of the function χ(i) ≡ λ~i-1~/δ~i~. Let us suppose that, for large i, the following expansion is valid: χ(i) ≡ λ~i-1~/δ~i~= i^s^θ(1-γ/i+ O(1/i^2^)) (1.5) Then, the asymptotical behavior of the stationary distribution of model (1.1) is completely defined by three parameters: s, θand γ(\[[@B12]\]). In particular, if the birth-and-death process is the 1^st^order balanced, i.e. if, by definition, s= 0 in (1.5), then, asymptotically, p~i~\~ θ^i^i^-γ^. If the process is 2^nd^order balanced, i.e. s= 0 and θ= 1, then p~i~\~ i^-γ^. The complete description of all possible asymptotics of the ergodic distributions of model (1.1) under condition (1.5) is given in Mathematical Appendix, Theorem 1 (hereinafter all references of the form (A.m.n) refer to the corresponding formula in the Mathematical Appendix \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]). It asserts that a large class of models, namely the second order balanced BDIMs, provide any given power asymptotic of the stationary frequency distributions of family sizes. 2. Classification of BDIMs -------------------------- ### Linear BDIM The simplest model that shows the generalized Pareto distribution is the linearBDIM with λ~i~= λ(i+a), δ~i~= δ(i+ b) for i\> 0, λ, δ, aand bare constants. (2.1) The equilibrium distribution of domain family sizes is defined by: ![](1471-2148-4-32-i5.gif) So, if λ= δ(θ= 1), the resulting 2^nd^order balanced linear BDIM has a power asymptotics with γ= 1 + b- a. ### Polynomial BDIM Informally, polynomial BDIMs can be introduced as follows. Under the linear BDIM, the dependence of the birth and death rates on family size is very weak; although each gene \"senses\" the size of the family (as reflected in the non-zero parameters aand b), this dependence cannot be interpreted as a specific form of interaction between family members. If such interactions are postulated, λ~i~\~ P~n~(i) and/or δ~i~\~ Q~m~(i), where P~n~(i), Q~m~(i) are polynomials on iof the n-th and m-th degrees. The ergodic distribution of the stochastic polynomial BDIM of the form (1.1) and (1.3) is asymptotically the same as that of the originally described deterministic polynomial BDIM \[[@B12]\], see Appendix (A.1.4), (A.1.5) \[see [Additional file 1](#S1){ref-type="supplementary-material"}\] and Proposition 2 for details. We show here that non-linear polynomial 2^nd^order balanced BDIM predict evolution rates that are dramatically greater than those for the linear BDIM. As an example, let us consider the quadratic BDIM in more detail. It takes into account the simplest, pairwise interaction between family members, which leads to λ~i~\~ i^2^and/or δ~i~\~ i^2^, i.e., one or both rates are polynomials on iof the second degree. If the polynomial degrees of the birth and death rates are different (e.g., λ~i~\~ iand δ~i~\~ i^2^), the corresponding BDIM is non-balanced, and equilibrium frequencies have no power asymptotics. Thus, let λ~i~= λ(i^2^+ r~1~i+ r~2~), δ~i~= δ(i^2^+ q~1~i+ q~2~), (2.3) where λ, δ, r~k~, q~k~, k= 1,2 are constants (such that λ~i~, δ~i~are positive for all i); equivalently, λ~i~= λ(i+ a)(i+ a~2~), δ~i~= δ(i+ b)(i+ b~2~). Then, r~1~= a+ a~2~, q~1~= b+ b~2~, and χ(i) = λ~i-1~/δ~i~= θ(1 + (r~1~- q~1~- 2)/i+ O(1/i^2^)), where θ= λ/δ. The quadratic BDIM has equilibrium sizes of domain families (see A.1.6) p~i~≈ c~2~p~0~λ~0~/λθ^i^i^ρ-2^ where ρ= r~1~- q~1~, c~2~= p~0~\[(Γ (1 + b) Γ (1 + b~2~)\] / \[Γ(1 + a) Γ (1 + a~2~)\], and ![](1471-2148-4-32-i6.gif) Thus, if the quadratic BDIM is 2^nd^order balanced, then p~i~\~ i^ρ-2^. Note that the asymptotic behavior frequencies p~i~do not depend on free coefficients r~2~and q~2~in (2.3), but only on θand r~1~- q~1~, although the constant c~2~could depend on the free coefficients r~2~, q~2~. ### Rational BDIM Rational models comprise a rather general class of BDIMs, for which the asymptotic behavior of the equilibrium frequencies and equilibrium sizes of domain families are fully tractable. The ergodic distribution of the stochastic rational BDIM is asymptotically the same as that of the deterministic rational BDIM \[[@B12]\]. In particular, if the model is 2^nd^order balanced, then p~i~\~ i^-γ^, (see A.1.2 and Proposition 1 in the Appendix for details \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]). The rational BDIMs can describe a substantially wider class of birth and death rates compared to polynomial models. In particular, birth rate can have a maximum at some specific value of family size and then decrease with further growth of the size, e.g., as shown in Fig. [1](#F1){ref-type="fig"}. This dependence of rates on family size can be described by the rational model with λ~i~= λ(i+ a~1~)/(i+ a~2~)^2^, δ~i~= δ(i+ b~1~)/(i+ b~2~)^2^. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Dependence of the birth rate (λ~i~= (i+ c~1~)/(i+ c~2~)^2^) on the family size. ::: ![](1471-2148-4-32-1) ::: ### Logistic BDIM Evidently, the number of size classes of protein families, N, should be finite, although intrinsic features that could determine the value of Nso far have not been considered (the impossibility of an infinite genome is self-evident but one would expect a much tighter bound based, e.g., on the limited time and resources available for genome replication and expression). Under the BDIMs described above, birth rate grows monotonically as the family size increases from 1 to Nand then abruptly drops to 0 (since families of size N+1 or greater are not allowed). However, this behavior is an arbitrary simplification of the model and hardly can reflect the actual process of genome evolution. In population dynamics models, the finiteness of a population size typically results from the \"saturation type\" of growth: the growth rate tends to 0 as the population size tends to the maximal possible value (see, e.g., \[[@B47]\]). It seems likely that, during genome evolution, gene duplication (and death) rate also tends to 0 as duplications leading to increase in gene number become deleterious when the size of some paralogous families becomes prohibitively large. The simplest formalism, which yields this type of population growth, is the logistic form of the birth rate. Logistic-like stochastic models have been investigated in various applications (e.g., \[[@B48],[@B49]\]), which considered a birth-and-death process with the rates λ(i) = c~3~(c~1~+ i)(N-i), δ(i) = c~3~i(c~2~-i), c~k~\> 0, k= 1,2,3, c~2~\>N. This model produces log-normal and log-series distributions; with the appropriate values of parameters, power low distributions of frequencies also appear, but only for intermediate values of i, namely, 1 \<\<i\<\<Nand N\>\> 1. ### Non-linear transformation of BDIM We have shown previously \[[@B43]\] that the following modification of any form of BDIM: λ\~i~= λ~i~g(i), δ\~i~= δ~i~g(i-1) (2.4) where g(i), i= 0,\...N, is a positive function, g(0) = 1, results in a BDIM with the same ergodic distribution of the family sizes as the original one. In particular, modifications of a linear BDIM with g(i) = (i+ 1)^d-1^or g(i) = (i+ 1)^d-1^(1 - i/(N+ c)) define, respectively, wide classes of rational or logistic BDIMs with the same stationary distribution as the original linear BDIM, but with manifestly different dynamic properties. 3. Probability of formation of a family of the given size before extinction and mean and variance of extinction time -------------------------------------------------------------------------------------------------------------------- In is known \[[@B44]\] that the probability for the birth-and-death process to reach state nbefore reaching state 0 from an initial state i\> 0 is given by formula (A.2.2). In terms of BDIM (1.1), this means that the probability of formation of a family of size nstarting from a family of size ibefore getting to extinction is given by (A.2.2). The random birth-and-death process (1.1) certainly visits state 0 in the course of time; this means that any domain family will eventually get extinct (and then, formally, can be \"reborn\", returning from the 0-class). Below we compute the mean time to extinction of a family of the given size for different versions of BDIM; the mean time to extinction of the largest family in the given genome is of particular interest. Let us denote S(n)=inf{t:X(t) = 0\\|X(0)=n} the time to the first passage of state 0 from the initial state n; S(n) is a random variable for each n. The mean time to extinction of the family of initial size n, E(S(n)), is given by the general formula A.3.2. ### Linear BDIM We have shown previously that, for the linear 2^nd^order balanced BDIM, the probability that a singleton expands to a family of size nbefore dying, P^(1)^(1,n) has the power asymptotics for large n(A.2.5). The values of probabilities P^(1)^(1,n) for different species are shown in Table [1](#T1){ref-type="table"}; these probabilities are no greater than \~10^-4^- 10^-5^. The mean time to extinction, E(S(n)), can be calculated using the relation E(S(n)) = 1/λE^(1)^~n~, where E^(1)^~n~, the mean time to extinction expressed in the 1/λtime units, is given by formula A.3.3 (see Table [1](#T1){ref-type="table"} for some numerical data and Figs. [1](#F1){ref-type="fig"},[2](#F2){ref-type="fig"} in \[[@B43]\]). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Family evolution under the linear BDIM (d=1) ::: ---------------------------------------------------------------------------------------------------------------------------------------------------- N P^(d)^(1,N) \10^2^* e^(d)^~N~ E^(d)^~N~ f^(d)^~N~ M^(d)^~N~ M^(d)^~N~/E^(d)^~N~ c^(d)^~du~ T^(d)^~N~\ ----- ------ ----------------------- -------------- ------------- -------------- ------------- ----------------------- -------------- -------------- Sce 130 0.284 295267 47.46 260080 20381.6 429.5 1.903 1939.3 Dme 335 0.227 778830 153.74 734725 37409.9 243.3 1.784 3337.0 Cel 662 0.160 1.866\10^6^ 347.76 1.803\10^6^ 68709.6 197.6 1.523 5232.2 Ath 1535 0.016 2.150\10^7^ 702.65 2.087\10^7^ 529639. 753.8 2.382 63080.0 Hsa 1151 0.026 1.329\10^7^ 505.26 1.29\10^7^ 300665. 595.1 2.721 40905.5 Tma 97 0.060 681356 31.47 513450 80677.3 2563.6 1.109 4473.6 Mth 43 1.125 37131.5 14.91 28570 4707.04 315.9 1.091 256.8 Sso 81 0.461 129115 30.14 98440 12853.5 426.5 1.253 805.3 Bsu 124 0.284 237343 48.89 202150 22921.0 468.8 1.320 1512.8 Eco 140 0.155 440665 51.67 375943 37959.8 734.7 1.544 2930.5 ---------------------------------------------------------------------------------------------------------------------------------------------------- For the linear BDIM (d= 1) and for the largest family of size Nin each genome, the table shows the probability of formation P^(d)^(1,N), mean number of events before extinction of the largest family e^(d)^~N~; mean number of events before formation of the largest family from a singleton, f^(d)^~N~; mean times of formation M^(d)^~N~and extinction E^(d)^~N~(in 1/λunits); the value of coefficient c^(d)^~du~= r~du~/λ; mean times of formation T^(d)^~N~in Ga (10^9^yrs) under r~du~= 2 × 10^-8^. The model parameters were genome-specific as determined previously \[12\]. and were the same for all model degrees according to (2.4). Species abbreviations: Sce, Saccharomyces cerevisiae, Dme, Drosophila melanogaster, Cel, Caenorhabditis elegans, Ath, Arabidopsis thaliana, Hsa, Homo sapiens, Tma, Thermotoga maritima, Mth, Methanothermobacter thermoautotrophicum, Sso, Sulfolobus solfataricus, Bsu, Bacillus subtilis, Eco, Escherichia coli. ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Coefficient of variation of the extinction time versus the family size for the linear BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. sapiens(red), A. thaliana(green) (Table 1 in \[43\]). ::: ![](1471-2148-4-32-2) ::: The variance of extinction time Var(S(n)) for the linear 2^nd^order balanced BDIM is Var(S(n)) = 1/λ^2^W^(1)^~n~, where W^(1)^~n~can be calculated using the formula (A.3.7). The plot of the coefficient of variation s^(1)^~n~= (W^(1)^~n~)^1/2^/E^(1)^~n~versus nfor different species is shown in Fig. [2](#F2){ref-type="fig"} (see also Table [1](#T1){ref-type="table"} for some numerical data). Clearly, the extinction time can vary within an extremely broad range of values. ### Non-linear polynomial and rational BDIM The stochastic behavior of the system and its characteristics also can be investigated within the broader framework of rational BDIMs. We will examine models represented as transformed linear BDIM (2.1), with λ~i~= λ(i+ a)(i+ 1)^d-1^, δ~i~= λ(i+ b)i^d-1^, (3.1) where d≥ 1 is the model degree. Let us recall that Theorem 1 (Mathematical Appendix \[see [Additional file 1](#S1){ref-type="supplementary-material"} \]) shows that the highest degrees and the corresponding coefficients of the birth and death rates at i^d^must be equal to provide for the power asymptotics of the stationary distribution, P(i) \~ i^-γ^. The power γof this distribution is completely determined by the degree dand the coefficients at i^d-1^. Thus, the model (1.1), (3.1) is representative of all rational BDIMs of the degree dwith a given power asymptotic (γ= b- a+ 1) of the stationary distribution. Besides, according to Proposition 1, this distribution for model (3.1) is exactly the same as for the corresponding linear model with λ~i~= λ(i+ a), δ~i~= λ(i+ b), which was studied in detail in \[12\]. We applied formula (A.2.6) ![](1471-2148-4-32-i7.gif) with g(i) = (i+ 1)^d-1^, to calculate the probability of formation of a family of the given size from a singleton before getting to extinction for the BDIM of degree d, P^(d)^(1,n). For example, the probabilities P^(2)^(1,n) and P^(3)^(1,n) for the quadratic and cubic BDIMs, respectively, are given by this formula with g(i) = i+ 1 and g(i) = (i+ 1)^2^, respectively. Figures [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"} show the dependence of the probabilities P^(2)^(1,n) and P^(3)^(1,n) on the family size nfor different species. The dependence of the probability P^(d)^(1,N) of the formation of the largest family on the model degree is shown in Fig. [5](#F5){ref-type="fig"}. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Probability of family formation starting from a singleton, P^(2)^(1,n), versus the family size (n) for the quadratic BDIM (in double logarithmic scale). The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Probability of family formation from a singleton, P^(3)^(1,n), versus the family size (n) for the cubic BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Probability of formation of the largest family starting from a singleton, P^(d)^(1,N), for rational BDIMs depending on the model degree d. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-5) ::: The mean time to extinction for the rational BDIM (1.1), (3.1) with a fixed dis calculated using the formula (A.3.4) ![](1471-2148-4-32-i8.gif) where ![](1471-2148-4-32-i9.gif) Here E^\^~n~is the mean time to extinction in the 1/λtime units. Figures [6](#F6){ref-type="fig"} and [7](#F7){ref-type="fig"} show the dependence of E^(2)^~n~and E^(3)^~n~on nfor the quadratic and cubic BDIMs, respectively. Fig. [8](#F8){ref-type="fig"} shows the mean times of extinction of the largest family, E^(d)^~N~, for different species, depending on the model degree d. Some numerical values of the mean time to extinction for quadratic and cubic BDIMs and different species are given in Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}. The variance of the extinction time of a family of size n, Var(S(n))= 1/λ^2^W^(d)^(n), d= 2, 3 for the quadratic and cubic BDIMs, and the coefficient of variation s^(d)^~n~= (W^(d)^~n~)^1/2^/E^(d)^~n~are calculated using the formulas (A.3.8). The results are shown in Figs. [9](#F9){ref-type="fig"} and [10](#F10){ref-type="fig"}. Some numerical values of the coefficient of variation of the extinction time for different species are given in Table [4](#T4){ref-type="table"}. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Mean time to extinction (in 1/λunits) depending on the family size for the quadratic BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Mean time to extinction (in 1/λunits) depending on the family size for the cubic BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Mean time to extinction (in 1/λunits) of the largest family for the rational BDIM depending on the model degree d. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Coefficient of variation of the time to extinction depending on the family size for the quadratic BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-9) ::: ::: {#F10 .fig} Figure 10 ::: {.caption} ###### The coefficient of variation of extinction time versus family size for the cubic BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-10) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Family evolution under the linear BDIM (d= 1) ::: ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- N* P^(d)^(1,N) \10^2^ e^(d)^~N~ E^(d)^~N~ f^(d)^~N~ M^(d)^~N~ M^(d)^~N~/E^(d)^~N~ c^(d)^~du~ T^(d)^~N~\ ----- ------ --------------------------- ----------------- ----------------- ----------------- ----------------- --------------------------------- ------------------ ------------------ Sce 130 0.230 33206.9 2.82 32772 249.80 88.58 7.56 94.4 Dme 335 0.404 127814. 4.72 127567 206.26 43.71 11.67 120.4 Cel 662 0.498 394794. 6.61 394593 215.36 32.58 15.80 170.2 Ath 1535 0.131 2.768\10^6^ 5.98 2.77\10^6^ 638.27 106.73 22.50 718.1 Hsa 1151 0.166 1.555\10^6^ 5.37 1.68\10^6^ 468.84 87.31 24.48 573.9 Tma 97 0.039 38872.6 2.25 36306 1231.3 547.26 3.27 201.3 Mth 43 0.315 4539.9 2.03 4234 166.47 77.09 3.33 27.7 Sso 81 0.233 13281.1 2.61 12852 252.47 97.11 4.33 54.7 Bsu 124 0.212 26441.0 3.10 25969 304.97 98.38 5.09 77.6 Eco 140 0.135 34970.6 2.90 40270 431.85 148.91 5.74 123.9 ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- For the quadratic BDIM (d= 2) and for the largest family of size Nin each genome, the table shows the probability of formation P^(d)^(1,N), mean number of events before extinction of the largest family e^(d)^~N~; mean number of events before formation of the largest family from a singleton,f^(d)^~N~; mean times of formation M^(d)^~N~and extinction E^(d)^~N~(in 1/λunits); the value of coefficient c^(d)^~du~= r~du~vλ; mean times of formation T^(d)^~N~in Ga (10^9^yrs) under r~du~= 2 × 10^-8^. The model parameters were the same as for the linear model according to (2.4). Species abbreviations: Sce, Saccharomyces cerevisiae, Dme, Drosophila melanogaster, Cel, Caenorhabditis elegans, Ath, Arabidopsis thaliana, Hsa, Homo sapiens, Tma, Thermotoga maritima, Mth, Methanothermobacter thermoautotrophicum, Sso, Sulfolobus solfataricus, Bsu, Bacillus subtilis, Eco, Escherichia coli. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Family evolution under the cubic BDIM (d= 3). ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ N P^(d)^(1,N) e^(d)^~N~ E^(d)^~N~ f^(d)^~N~\ M^(d)^~N~ M^(d)^~N~/E^(d)^~N~ c^(d)^~du~= r~du~vλ* T^(d)^~N~ ------ ------ ------------------- ----------------- ----------------- ------------------ ----------------- --------------------------------- --------------------------------- ----------------- Sc e 130 0.105 12315.7 0.944 12306 4.60 4.84 92.46 21.3 Dme 335 0.222 60759.4 1.390 60755 2.45 1.76 549.65 67.3 Cel 662 0.283 208472 1.804 208469 2.10 1.17 2020.37 212.1 Ath 1535 0.255 1.29\10^6^ 1.390 1.29\10^6^ 1.93 1.39 3754.83 362.3 Hsa 1151 0.254 756242 1.291 756238 1.65 1.27 2938.07 242.4 Tma 97 0.019 9442.5 0.781 9390 24.5 31.4 18.84 23.1 Mth 43 0.061 1530.2 0.848 1514 7.85 9.24 18.26 7.2 Sso 81 0.073 4799.6 0.960 4786 7.21 7.51 36.71 13.2 Bsu 124 0.088 10265.3 1.059 10254 6.40 6.04 63.38 20.3 Eco 140 0.071 14459.9 0.957 14446 7.34 7.67 65.06 23.9 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ For the cubic BDIM (d= 3) and for the largest family of size Nin each genome, the table shows the probability of formation P^(d)^(1,N), mean number of events before extinction of the largest family e^(d)^~N~; mean number of events before formation of the largest family from a singleton, f^(d)^~N~; mean times of formation M^(d)^~N~and extinction E^(d)^~N~(in 1/λunits); the value of coefficient c^(d)^~du~= r~du~vλ; mean times of formation T^(d)^~N~in Ga (10^9^yrs) under r~du~= 2 × 10^-8^. The model parameters were the same as for the linear model according to (2.4). Species abbreviations: Sce, Saccharomyces cerevisiae, Dme, Drosophila melanogaster, Cel, Caenorhabditis elegans, Ath, Arabidopsis thaliana, Hsa, Homo sapiens, Tma, Thermotoga maritima, Mth, Methanothermobacter thermoautotrophicum, Sso, Sulfolobus solfataricus, Bsu, Bacillus subtilis, Eco, Escherichia coli. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Coefficients of variation of the extinction and formation times for the BDIMs of different degrees ::: ------------------------------------------------------------------------------------------------------------- N s^(1)^~N~\ σ^(1)^~N~ s^(2)^~N~ σ^(2)^~N~ s^(3)^~N~ σ^(3)^~N~ ----- ------ ---------------- --------------- --------------- --------------- --------------- --------------- Dme 335 194.11 81.79 304.96 126.90 766.29 184.70 Cel 662 413.30 195.73 460.31 277.24 481.65 391.25 Ath 1535 885.78 421.03 1016.85 583.56 1042.86 886.95 Hsa 1151 649.77 308.40 746.56 425.21 768.04 647.23 ------------------------------------------------------------------------------------------------------------- The table shows coefficient of variation s^(d)^~N~of extinction timefor the largest family; coefficient of variation σ^(d)^~N~of formation time for the largest family; d= 1,2,3 for the linear, quadratic and cubic BDIM, respectively. Species abbreviations: Dme, Drosophila melanogaster, Cel, Caenorhabditis elegans, Ath, Arabidopsis thaliana, Hsa, Homo sapiens. ::: ### Logistic BDIM Let us consider the logistic modification of the rational BDIM; specifically, we will examine models with the birth and death rates of the form λ~i~= λ(i+ a)(i+ 1)^d-1^(1 - i/(N+ c)), δ~i~= δ(i+ b)i^d-1^(1 - (i- 1)/(N+ c)). (3.2) We will refer to the parameter cas saturation boundary. The shape of λ~i~essentially depends on the value of c(Fig. [11](#F11){ref-type="fig"}). ::: {#F11 .fig} Figure 11 ::: {.caption} ###### Dependence of λ~i~(3.2) at d= 2 on iwith different boundary value, c= 1, c= 100, c= 1000 (from bottom to top). The model parameters are for Drosophila melanogaster. ::: ![](1471-2148-4-32-11) ::: The logistic model (1.1), (3.2) is a transformation (2.9) of the linear BDIM using the function: g(i) = (i+ 1)^d-1^(1 - i/(N+ c), c= const ≥ 0. (3.3) The stationary distribution of family size frequencies for the logistic model (1.1), (3.2) is exactly the same as that for corresponding linear BDIM but the stochastic properties are different and close to the rational models, and essentially depend on the boundary c. With a large c, the model is very close to the corresponding rational model with λ~i~= λ(i+ a)(i+ 1)^d-1^, δ~i~= δ(i+ b)i^d-1^, but with small c, we can observe some new effects when the family size approaches N. The probability of formation of a family of a given size from a singleton before getting to extinction for the logistic BDIM is calculated using the general formula (A.2.6) where the function g(i) is given by (3.3). The dependence of this probability on the model degree dunder a fixed large value of the boundary c\~Nis similar to that for the corresponding rational models but differs under a small c; Fig. [12](#F12){ref-type="fig"} shows this dependence for c= 1. ::: {#F12 .fig} Figure 12 ::: {.caption} ###### Dependence of the probability P^(d)^(1,n) on the family size nfor the logistic model with c= 1 for d= 1, 2 and 3 (from bottom to top). The model parameters are for Drosophila melanogaster. ::: ![](1471-2148-4-32-12) ::: The mean times of extinction for the logistic BDIMs are calculated using formula (A.3.4). Fig. [13](#F13){ref-type="fig"} shows the mean times of extinction of the largest family, E^(d)^~N~, depending on the model degree dfor different values of saturation boundary c. Fig. [14](#F14){ref-type="fig"} shows the dependence of E^(d)^~N~on the saturation boundary cfor different values of d. ::: {#F13 .fig} Figure 13 ::: {.caption} ###### Mean time to extinction (in 1/λunits) of the largest families for the logistic BDIM depending on the model degree dfor c= 1,c= 100 and c= 1000 (from top to bottom, in double logarithmic scale). The model parameters are for Drosophila melanogaster. ::: ![](1471-2148-4-32-13) ::: ::: {#F14 .fig} Figure 14 ::: {.caption} ###### Mean time to extinction (in 1/λunits) of the largest families for the logistic BDIM, dependently on the boundary value cfor d= 1 (left) and d= 2 (right). The model parameters are for Drosophila melanogaster. ::: ![](1471-2148-4-32-14) ::: 4. Mean and variance of formation time for a family of the given size --------------------------------------------------------------------- Let us denote T(j, n) = inf{t: X(t) = n\\|X(0) = j} the time to the first passage of state nfrom the initial state j; T(j, n) is a random variable for each j, n. The mean time to the first passage for BDIM (1.1), m(j, n) = E(T(j, n)), can be calculated using the formula m(j, n) = m~0~(j, n) + m~1~(j, n). Here the term m~0~(j,n) is the mean time elapsed before the system leaves the 0 state for the last time, and the term m~1~(j,n) is the mean time of formation of a family of size nfrom a singleton after its last \"resurrection\" (see formulas (A.4.1) for details). Below we examine only the mean family formation time from an essential singleton (model (1.3)). ### Linear BDIM Previously, we determined the mean time of formation of a family of size nfrom a singleton for different species \[[@B43]\]. For the linear BDIM, the value of the mean formation time from an essential singleton is given by formula M^(1)^(1,n) = 1/λM^(1)^~n~, where M^(1)^~n~, the mean formation time in 1/λunits is calculated using the formula (A.4.6) ![](1471-2148-4-32-i10.gif) The transition from the 1/λtime units to years is considered in s.6 of the Mathematical Appendix \[see [Additional file 1](#S1){ref-type="supplementary-material"} \]. The mean formation time E(T(1, n)) in years is calculated using the formula (A.6.4) and the current empirical estimates of the gene duplication rate \[[@B24]\]. Plots of E(T(1, n)) for different species are shown in Fig. [15](#F15){ref-type="fig"}. ::: {#F15 .fig} Figure 15 ::: {.caption} ###### Mean time to formation (in years, Ga, with r~du~= 2\10^-8^) depending on family size for the linear BDIM (double logarithmic scale). The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-15) ::: Once we computed the meantime of formation of a family of size nfor different species, the question arises how accurately is the time T(i, n) of the first random passage through the threshold npredicted by the mean value. To address this problem, we estimated the variance of the time of family formation, Var(T(i, n)) using the general formulas (A.5.2) for model (1.1) and (A.5.3) for model (1.3), respectively. For the linear BDIM, the variance of the formation time for a family of size nfrom an essential singleton, V^(1)^~n~, is given by the formula (A.5.5). A more important and informative characteristic, which is independent on the model parameter λ, is the coefficient of variation, which is equal to ![](1471-2148-4-32-i11.gif). The coefficient of variation of the formation time of a family of size nfrom a singleton, σ^(d)^~n~= (V^(d)^~n~)^1/2^/M^(d)^(1;n) for the BDIM of degree dis the most relevant value. The plots of σ^(1)^~n~versus nfor the linear model and for different species are shown in Fig. [16](#F16){ref-type="fig"}. ::: {#F16 .fig} Figure 16 ::: {.caption} ###### The coefficient of variation σ^(1)^~n~of family formation time depending on nfor the linear BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-16) ::: The coefficients of variation were very large for all species (see numerical values in Table [4](#T4){ref-type="table"}). To summarize the results obtained for the stochastic characteristics of the linear BDIM, we found that: i) under this model, the mean time to extinction of the largest families in most genomes was much shorter than the mean time of formation of these families, and ii) using the current estimates of duplication rates in eukaryotic genomes (r~du~≈ 2 × 10^-8^duplications/gene/year \[[@B24]\]) to express the mean family formation times in real time units instead of the dimensionless 1/λunits, we obtain M^(1)^(1;N) \~ 10^13^- 10^14^yrs, a completely unrealistic time estimate. The mean family formation times given by the linear BDIM would become realistic only if the recent analyses underestimated the gene duplication rate by a factor of \~10^4^, which does not seem plausible. Thus, the linear BDIM cannot provide an adequate description of genome evolution, at least when only the mean time of family formation is considered. The variance of the family formation time is extremely large (the coefficient of variation is \~10^2^), and, accordingly, large deviations from the mean time, more to two orders of magnitude, are possible. However, even taking this into account, the family formation times predicted by the linear BDIM are far longer than the time allotted for life\'s evolution of earth. In the next section, we consider non-linear, higher order models that have the potential to yield shorter mean times of family formation. ### Polynomial BDIMs The mean time of formation of families from an essential singleton (or after the last \"resurrection\" of a family) depending on the family size nfor the polynomial BDIMs is E(T(1,n)) = 1/λM\~n~where M\~n~, the mean time of formation in 1/λunits can be calculated using the formulas (A.4.9) ![](1471-2148-4-32-i12.gif) Fig. [17](#F17){ref-type="fig"} shows the dependence of the mean time of family formation on the family size for the quadratic BDIM in years, calculated using the formula (A.6.4). The values of mean times of formation for this BDIM are given in Table [2](#T2){ref-type="table"}. ::: {#F17 .fig} Figure 17 ::: {.caption} ###### Mean time of formation (in years, Ga, with r~du~= 2\10^-8^) depending on family size nfor the quadratic BDIM (in double logarithmic scale). The model parameters are for D. melanogaster(blue),C. elegans(purple),H. Sapiens(red),Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-17) ::: The variance of formation time of a family of the size ncan be calculated using the formula (A.5.6), with g(j)=j+1 for the quadratic BDIM and g(j) = (j+ 1)^2^for the cubic BDIM, respectively. The dependence of the coefficient of variation σ^(2)^~n~= (V^(2)^(1,n))^1/2^/M^(2)^(1;n) on the family size for the quadratic BDIM is shown in Fig. [18](#F18){ref-type="fig"}, and some numerical data are given in Table [4](#T4){ref-type="table"}. ::: {#F18 .fig} Figure 18 ::: {.caption} ###### The coefficient of variation σ^(2)^~n~of formation time versus family size for the quadratic BDIM and species. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-18) ::: Although the variance of family formation times for the quadratic BDIM is approximately 5 orders of magnitude less than that for the linear BDIM, the values of the coefficient of variation for quadratic BDIM are about 1.3--1.5 times greater than those for the linear BDIM. Thus, the actual formation time for the largest family could differ from the mean value by several orders of magnitude with a high probability. Figures [19](#F19){ref-type="fig"} and [20](#F20){ref-type="fig"} show the dependence of the mean and the coefficients of variation of family formation time on family size for the cubic BDIM. ::: {#F19 .fig} Figure 19 ::: {.caption} ###### Mean time of formation (in years, Ga, with r~du~= 2\10^-8^) depending on family size nfor the cubic BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-19) ::: ::: {#F20 .fig} Figure 20 ::: {.caption} ###### The coefficient of variation σ^(3)^~n~of formation time versus family size for the cubic BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-20) ::: We have shown previously that the cubic model shows extremely high evolution rate comparatively with the linear and even quadratic models under the same value of the parameter λ\[[@B43]\]. On the contrary, the mean formation times in years for the quadratic and cubic models are of the same order (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). The polynomial models bring the mean time required for the formation of families of the observed size closer to realistic values but these times still remain far too long. Specifically, with the empirical estimates of the duplication rates used above for the linear BDIM, the quadratic model gives the mean family formation times \~10^11^yrs. This value is close to the minimum possible time of family formation that can be calculated using the duplication rate estimates of Lynch and Conery \[[@B24]\] and non-linear rational BDIMs. ### Non-linear rational BDIMs Let us investigate the dependence of the dynamics of the mean time of family formation on the model degree and the family size. The mean time of formation of a family of size nfrom a singleton under a fixed model degree d, M^(d)^(1;n), for the rational BDIM (1.1),(3.1), is calculated using the formula (A.4.9). A comparison of the mean time of formation and extinction for rational BDIMs reveals an interesting property of non-linear BDIMs: for any given family size n, there exists such a model degree that the times of family formation and extinction are equal (as it becomes obvious from the intersection of the respective curves in Fig. [21](#F21){ref-type="fig"}). Accordingly, at higher model degrees, the mean time of formation becomes shorter than the mean time to extinction. The model degree that corresponds to the point of intersection in Fig. [21](#F21){ref-type="fig"} obviously depends on the size of the considered family. Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"} show that the mean time of formation is about 100 times more than the mean time to extinction for the largest families of different species for the quadratic BDIM and only about 10 times more for the cubic model. ::: {#F21 .fig} Figure 21 ::: {.caption} ###### Mean times (in 1/λunits) of formation (upper curve beforethe point of intersection) and extinction (upper curve afterthe point of intersection) of the largest family depending on the model degree (semi-logarithmic scale). The model parameters are for Homo sapiens.* ::: ![](1471-2148-4-32-21) ::: As shown previously, increasing the degree (the \"order of interaction\") dresults in indefinite decrease of the family formation time expressed in 1/λtime units (\[[@B43]\] and Fig. [22](#F22){ref-type="fig"}). However, we have also shown that this effect is offset by the rapid increase of the average duplication rate in the model. Assuming the gene duplication rate of \~2\10^-8^year^-1^\[[@B24]\], the evolution time in years, calculated according to the formula (A.6.5), does not decrease indefinitely, but has a minimum at the model degree between 2 and 3 (Fig. [23](#F23){ref-type="fig"}). Even the minimum mean time of the largest family formation achievable with the rational BDIMs is on the order of 10^11^years (see Table [6](#T6){ref-type="table"}), which is incompatible with the age of life on Earth \[[@B43]\]. Thus, a rational BDIM of any degree cannot provide an adequate description of genome evolution, at least when only the mean time of family formation is considered. Accordingly, for assessing the feasibility of the formation of the largest families under a given model, the variance of the formation time should be investigated. ::: {#F22 .fig} Figure 22 ::: {.caption} ###### Mean time of formation of the largest family (in 1/λunits), M^(d)^~N~, for the rational BDIM depending on the model degree d(double logarithmic scale). The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-22) ::: ::: {#F23 .fig} Figure 23 ::: {.caption} ###### Dependence of the time (in years, Ga) required for the formation of the largest family on the model degree dfor the rational BDIM (semi-logarithmic scale). The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-23) ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Rational BDIM yielding the shortest mean time of family formation ::: N* D R^(D)^(N) T^(D)^~N~ ----- ------ ------ ----------------- ----------------- Sce 130 3.13 416.0 20.8 Dme 335 2.67 1131.0 56.55 Cel 662 2.44 2317.7 115.9 Ath 1535 2.65 5553.8 277.7 Hsa 1151 2.71 4079.5 204. Tma 97 3.56 317.8 15.9 Mth 43 2.40 125.2 6.3 Sso 81 2.19 254.2 12.7 Bsu 124 2.05 404.4 20. Eco 140 2.16 460.4 23. For each genome, D is the value of the model degree d, which results in the minimum of the mean time of formation of the largest family, T^(d)^~N~= R^(d)^(N)/r~du~(in Ga, under indicated value of dand r~du~= 2 × 10^-8^) are shown. Species abbreviations: Sce, Saccharomyces cerevisiae, Dme, Drosophila melanogaster, Cel, Caenorhabditis elegans, Ath, Arabidopsis thaliana, Hsa, Homo sapiens, Tma, Thermotoga maritima, Mth, Methanothermobacter thermoautotrophicum, Sso, Sulfolobus solfataricus, Bsu, Bacillus subtilis, Eco, Escherichia coli. ::: Generally, the variance of the formation time of the family of the given size is given by the formulas (A.5.3) and (A.5.6). Although the variance of formation times for the quadratic and, especially, for the cubic BDIM is several orders of magnitude less than that for the linear BDIM, the coefficients of variation for both formation and extinction time increase with the model degree (Table [4](#T4){ref-type="table"}). These coefficients are so large that the actual formation time of the largest family could differ from its mean value by several orders of magnitude with a high probability. ### Logistic BDIM The mean time of formation (in 1/λunits) of a family of size nfrom an essential singleton for the logistic BDIM (1.3), (3.2) under fixed dis calculated using formula (A.4.9). Fig. [24](#F24){ref-type="fig"} shows the dependence of mean times of family formation, M^(d)^(1;n), on the family size nfor different model degrees dunder the fixed saturation boundary c= 1, and Fig. [25](#F25){ref-type="fig"} shows the dependence of mean times of family formation on the boundary value (see Tables [7](#T7){ref-type="table"} and [8](#T8){ref-type="table"} for some numerical data). Similarly to the rational BDIM, increasing the degree (the \"order of interaction\") of the logistic model results in faster family evolution under a fixed value of the parameter λ. However, when this inner model parameter is excluded and the mean time of family formation is expressed in yearsaccording to formula (A.6.5), then we again face a restriction that does not allow indefinite shortening of the family formation time, T^(d)^~N~. Specifically, T^(d)^~N~for the logistic model with a fixed Nhas a minimum over d. We identified the model degrees yielding the minimum mean time of formation of the largest family for the logistic-rational BDIM. Fig. [26](#F26){ref-type="fig"} and Table [9](#T9){ref-type="table"} show the dependence of T^(d)^~N~on dfor the logistic model with fixed saturation boundary. ::: {#F24 .fig} Figure 24 ::: {.caption} ###### Mean time of formation (in 1/λunits) of a family of the given size depending on the size for the logistic BDIM with the boundary value c= 1 for d= 1, d= 2, d= 3 (from top to bottom, semi-logarithmic scale). The model parameters are for Drosophila melanogaster.. ::: ![](1471-2148-4-32-24) ::: ::: {#F25 .fig} Figure 25 ::: {.caption} ###### Mean time of formation (in 1/λunits) of the largest family for the logistic BDIM, depending on the model degree dfor c= 1, c= 100 and c= 1000 (from top to bottom, double logarithmic scale). The model parameters are for Drosophila melanogaster. ::: ![](1471-2148-4-32-25) ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Evolution of gene families under the logistic BDIM with c= 1 and different d. ::: ------------------------------------------------------------------------------------------------------------------------------------------------------- P^(d)^(1,N) E^(d)^~N~ M^(d)^~N~ M^(d)^~N~/ E^(d)^~N~ c^(d)^~du~= r~du~vλ T^(d)^~N~\ ------- ------------------- ----------------- ----------------- ---------------------------------- --------------------------------- ------------------ d = 1 0.24\10^-7^ 314.72 351042. 1115.4 1.7545 30795.2 d = 2 0.68\10^-3^ 5.66 1247.3 220.37 10.073 628.20 d = 3 0.113 1.41 6.14 4.35 297.29 91.27 ------------------------------------------------------------------------------------------------------------------------------------------------------- Model parameters are for D. melanogaster. ::: ::: {#T8 .table-wrap} Table 8 ::: {.caption} ###### Evolution of gene families under the logistic BDIM with c= 100 and different d. ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------- P^(d)^(1,N) E^(d)^~N~ M^(d)^~N~ M^(d)^~N~/E^(d)^~N~ c^(d)^~du~= r~du~vλ T^(d)^~N~ L^(d)^\ ------- ------------------- ----------------- ----------------- --------------------------------- --------------------------------- ----------------- ------------- d = 1 0.94\10^-5^ 227.19 90107.4 396.62 1.7612 7934.9 32.62 d = 2 0.2\10^-2^ 5.24 412.45 78.71 10.437 215.24 193.34 d = 3 0.178 1.40 3.39 2.42 354.72 25.25 6571.04 ------------------------------------------------------------------------------------------------------------------------------------------------------------------- Model parameters are for D. melanogaster. ::: ::: {#T9 .table-wrap} Table 9 ::: {.caption} ###### Logistic BDIM yielding the shortest mean time of family formation under c= 1 ::: N D R^(D)^(N) T^(D)^~N~ ----- ------ ------ ----------------- ----------------- Dme 335 3.18 1726.8 86.34 Cel 662 2.92 3749.5 187.5 Ath 1535 3.11 10234.5 511.7 Has 1151 3.19 7433.9 371.7 For each genome, Dis the value of model degree d, which results in the minimum of the mean time of formation of the largest family, T^(d)^~N~= R^(d)^(N)/r~du~(in Ga), is indicated. Species abbreviations: Dme, Drosophila melanogaster, Cel, Caenorhabditis elegans, Ath, Arabidopsis thaliana, Hsa, Homo sapiens. ::: ::: {#F26 .fig} Figure 26 ::: {.caption} ###### Dependence of the mean time (in years, Ga) required for the formation of the largest family for the logistic BDIM under fixed saturation boundary c= 1 on the model degree d(semi-logarithmic scale). The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-26) ::: Thus, as in the case of rational BDIMs, increase of the degree of logistic BDIMs under a fixed value of average duplication rate r~du~cannot yield mean family formation times \< 10^11^years. Furthermore, the \"saturation effect\" seen in the logistic models increases the mean time of family formation compared to the corresponding rational models (compare Tables [5](#T5){ref-type="table"} and [7](#T7){ref-type="table"}). ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Coefficients of variation of the number of events before formation of the largest family for the BDIMs of different degrees ::: N Σ^(1)^~N~ Σ^(2)^~N~ Σ^(3)^~N~ ----- ------ --------------- --------------- --------------- Dme 335 87.00 86.60 79.91 Cel 662 177.99 168.73 154.81 Ath 1535 402.66 399.03 366.50 Hsa 1151 296.42 299.31 276.23 Coefficient of variation Σ^(d)^~N~of the number of eventsbefore formation of the largest family; d= 1 for the linear BDIM, d= 2 for the quadratic BDIM, d= 3 for the cubic BDIM. Species abbreviations: Dme, Drosophila melanogaster, Cel, Caenorhabditis elegans, Ath, Arabidopsis thaliana, Hsa, Homo sapiens. ::: 5. The mean number of elementary events before family extinction and formation ------------------------------------------------------------------------------ Comparing the mean family formation and extinction times predicted by BDIMs with the actual evolutionary timescale allow us to choose the most appropriate version from the examined class of models. The numberof elementary evolutionary events namely, duplication and deletion of domains, predicted by these models is of potential interest in itself as an approximation of an important characteristic of genome evolution. To calculate the mean number of elementary events during evolution of gene families, we employed the so-called embeddingchains {Y(n)} instead of the original BDIM. The embedding chain {Y~n~} for a particular BDIM is a random walk with discrete time on the same set of states and transition probabilities p~i,i+1~= β~i~= λ~i~/(λ~i~+ δ~i~), p~i,i-1~= μ~i~= δ~i~/(λ~i~+ δ~i~) and p~ij~= 0 for all other cases (see s.7 of Mathematical Appendix for details \[see [Additional file 1](#S1){ref-type="supplementary-material"} \]). The transition from the state ito the state i+1 (or i-1) corresponds to the duplication (or deletion) of a domain in a family of size i. The only difference between the original birth-and-death process and the embedding chain is that the sojourn time for the embedding chain is equal to 1 for any state iinstead of 1/(λ~i~+ δ~i~). The ratio β~i~/μ~i~(= λ~i~/δ~i~) characterizes the trend of family evolution from the state i, i.e., is the family more likely to grow or to shrink; for a symmetricrandom walk, β~i~/μ~i~= 1 for all i. The dependence of the ratio β~i~/μ~i~on ifor different rational and logistic embedded chains is shown in Figures [27](#F27){ref-type="fig"} and [28](#F28){ref-type="fig"}. For the rational models, β~i~/μ~i~≈ 1 for large i; for the logistic models, β~i~/μ~i~≈ 1 for 0 \<\<i\<\<N(however, this ratio significantly deviates from 1 at both ends of the interval of states). Thus, the behavior of the embedding chain is similar to the behavior of the symmetric random walk in the corresponding subsets of states. Informally, the plots in Figures [27](#F27){ref-type="fig"} and [28](#F28){ref-type="fig"} indicate that small families may preferentially grow (under higher degree models) or shrink (under low degree models) whereas the evolution of large families tends to a symmetrical random walk. ::: {#F27 .fig} Figure 27 ::: {.caption} ###### The ratio β~i~/μ~i~against family size ifor the rationalBDIM depending on the model degree d:d= 1, d= 1.6, d= 2 (from bottom to top), in double logarithmic scale. The model parameters are for Drosophila melanogaster. ::: ![](1471-2148-4-32-27) ::: ::: {#F28 .fig} Figure 28 ::: {.caption} ###### The ratio β~i~/μ~i~against family size ifor the logisticBDIM (3.2) with c= 1 depending on the model degree d:d= 1, d= 1.6, d= 2 (from bottom to top). The model parameters are for Drosophila melanogaster ::: ![](1471-2148-4-32-28) ::: The mean number of elementary events before the formation of a family of the given size, f~n~, is computed using formulas (A.7.5)-(A.7.7). The plots in Figures [29](#F29){ref-type="fig"} and [30](#F30){ref-type="fig"} show the dependence of f~n~on the family size for different species for the linear and quadratic models, respectively. The mean number of elementary events before the extinction of a family of the given size, e~n~, is computed using formulas (A.7.13)-(A.7.15) and Figures [31](#F31){ref-type="fig"} and [32](#F32){ref-type="fig"} show the corresponding dependences for family extinction. Some numerical data for the mean number of elementary events for polynomial BDIMs are shown in Tables [1](#T1){ref-type="table"},[2](#T2){ref-type="table"},[3](#T3){ref-type="table"} and, for coefficients of variation, in Table [5](#T5){ref-type="table"}. Given that all the analyzed BDIMs are balanced, i.e., the birth and death rates are asymptotically equal, it was not unexpected that the mean number of events required for the formation of a large family (or the number of events preceding the extinction of such a family) was orders of magnitude greater than the size of the family. This suggests a highly dynamic picture of genome evolution whereby numerous duplications counterbalanced by gene losses are typically involved in the evolution of large families. However, the number of events required for the formation of a family of the given size quickly drops with the increase of a model degree (Fig. [33](#F33){ref-type="fig"}), which may be construed as reflection of positive selection leading to amplification of family members. ::: {#F29 .fig} Figure 29 ::: {.caption} ###### Mean number of events before the formation of a family of the given size for the linear BDIM (double logarithmic scale). The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-29) ::: ::: {#F30 .fig} Figure 30 ::: {.caption} ###### Mean number of events before the formation of a family of the given size for the quadratic BDIM (double logarithmic scale). The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-30) ::: ::: {#F31 .fig} Figure 31 ::: {.caption} ###### Mean number of events before extinction of a family of the given size for the linear BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-31) ::: ::: {#F32 .fig} Figure 32 ::: {.caption} ###### Mean number of events before extinction of a family of the given size for the quadratic BDIM. The model parameters are for D. melanogaster(blue), C. elegans(purple), H. Sapiens(red), Arabidopsis thaliana(green). ::: ![](1471-2148-4-32-32) ::: ::: {#F33 .fig} Figure 33 ::: {.caption} ###### Mean number of events before the formation of the largest family against the model degree for the rational BDIM (double logarithmic scale). The model parameters are for Drosophila melanogaster ::: ![](1471-2148-4-32-33) ::: 6. Monte Carlo simulation of evolution of gene family ensembles under BDIMs of different degrees ------------------------------------------------------------------------------------------------ As noticed previously \[[@B43]\], it is the minimum rather than the mean evolution time that is important for modeling the dynamics of evolution of genomes consisting of many gene families. Due to the large variance of the family formation time estimates (see the detailed discussion above), this value is likely to be much less than the mean. Although an analytical solution to this problem is hard to obtain, it can be examined in detail by Monte Carlo simulation analysis. As described previously \[[@B43]\], we employed for this analysis model parameters estimated for the human proteome. The simulated evolution started from 3000 families of size one (singletons) and continued until the largest family reached 1024 members (a convenient arbitrary number to approximate the size of the largest family in eukaryotic genomes); the simulation was run from 10 to several hundred times depending on the model degree (the time required for the simulation showed a complex, non-linear dependence on the model degree). In the course of the simulation, the number of families fluctuated due to stochastic births, deaths, and innovations of genes but, generally, tended toward the equilibrium number of \~1700, which is close to the empirically determined number of families in the human genome and is pre-determined by the choice of model parameters (the initial number of singletons did not have much impact on the model\'s dynamics). The time scale was adjusted such that r~du~= 2 × 10^-8^duplications/gene/year \[[@B24]\]. A series of simulations was performed for non-linear rational BDIMs with different degrees d. As shown in Fig. [34](#F34){ref-type="fig"}, the time at which the family size of 1024 members is reached for the first time depends on din a similar fashion as the mean time for a single family, i.e., there is clear minimum at a particular value of d. At the optimal value of d≈ 2.2, the model reaches this family size in 2.2 ± 0.5 Ga, which is comparable to the time of evolution of eukaryotes. Compared to the minimal evolution time predicted by BDIMs of different degrees for a single family, the genome-size ensemble of gene families reached the threshold size much faster (by 1.5--2.5 orders of magnitude), and the optimum values of dwas lower by \~0.5 (Fig. [35](#F35){ref-type="fig"}). The much faster formation of large families from an ensemble of singletons was predictable due to the large variation coefficient of the family formation and extinction times, but the simulation was necessary in the absence of knowledge of the exact distribution of these values. ::: {#F34 .fig} Figure 34 ::: {.caption} ###### The time required for the formation of a first family with 1024 members determined by Monte Carlo simulation starting from an ensemble of 3000 singletons. The model parameters are for Homo sapiens. ::: ![](1471-2148-4-32-34) ::: ::: {#F35 .fig} Figure 35 ::: {.caption} ###### The time required for the formation of a first family with 1024 members starting from an ensemble of 3000 singletons (blue) compared to the mean time predicted by BDIMs of different orders (magenta). The model parameters are for Homo sapiens. ::: ![](1471-2148-4-32-35) ::: 7. General discussion --------------------- Here and in the previous publications \[[@B12],[@B43],[@B50]\], we describe a general class of models, which are based on the classical concept of a birth-and-death process and seem to naturally apply to the genome evolution process. Similar, although not identical and apparently less general, modeling approaches have been considered by others \[[@B6],[@B34],[@B51]\]. Even earlier, evolution of gene families has been modeled within the distinct mathematical framework of multiplicative processes \[[@B52]\]. The utility of birth-and-death type models in evolutionary genomics in itself is not a trivial matter but rather stems from fundamental features of genome evolution. As captured in the title of Ohno\'s famous book \[[@B16]\], although foreseen even in the early days of genetics \[[@B15],[@B53]\], gene duplication probably is the principal mechanism of genome evolution. Of course, genomes cannot grow ad infinitumand, through most of the evolutionary history, the number of genes within a given phylogenetic lineage probably remains roughly constant. Hence duplication is intrinsically coupled to gene loss. The results of comparative genomics further show that many genes in each lineage cannot be obviously linked to other genes through duplication. Without necessarily specifying the biological mechanisms (these could involve rapid change after duplication, gene acquisition via horizontal transfer, and possibly, birth of genes from non-coding sequences), it is reasonable to view these unique genes as resulting from innovation. For genomes to maintain equilibrium, the combined rates of duplication and innovation over the entire ensemble of gene families should equal the rate of gene loss, at least when averaged over long time spans. The observed distribution of family sizes, which asymptotically tends to a power law, dictates a much more specific connection between the gene birth and death rates, namely, the second order balance. It should be noted that this form of balance does not amount to particularly fine tuning of the gene birth and death rates. The only requirement is that these rates tend to the same value when the family size tends to infinity according to the condition (1.5). In contrast, for small families, the rates may substantially differ, without significantly changing the shape of equilibrium distribution. The incentive to examine BDIMs in detail stems from at least two fundamental questions: i) are the above elementary evolutionary mechanisms sufficient to account for the empirically observed characteristics of genomes, ii) what is the contribution of natural selection to the general, quantifiable features of genomes, such as the size distribution of gene families. The analysis of BDIMs starts to provide some answers, albeit preliminary ones. The critical observation made in the course of BDIM analysis was that different versions of these models could be readily distinguished on the basis of goodness of fit to the empirical data. This being the case, we found that the simplest possible model, in which all paralogs are considered independent, is incompatible with the data. Thus, turning to the first of the above questions, we had to conclude that, in addition to the three elementary processes, \"something else\" was required to model genome evolution. This \"something\" is the dependence or \"interaction\" between gene family members which results in self-accelerating family growth. In order to account for the observed stationary distribution of family sizes, it is sufficient to introduce a very weak dependence as embodied in the linear BDIM. However, when we switched from the deterministic to the stochastic version of BDIMs, which provide for the possibility of analysis of the dynamics of the systems evolution, we found that evolution under the linear BDIM was much too slow to account for the emergence of the large families of paralogs found in all genomes during the time of life\'s evolution. Only higher order BDIMs, with degrees between 2 and 3, i.e., with \"strong interactions\" between family members were found to provide for sufficiently fast evolution to be compatible with the real biological timescale. Obviously, these findings beg the question: what is the nature of the mysterious \"interactions\" between paralogs? Although, on some occasions, paralogous protein do form physical complexes or interact functionally, the situation when such interaction does not exist is much more common. Therefore, the \"interactions\" in our models should not be perceived literally. This brings us to the second of the above major problems. BDIMs do not explicitly include the notion of selection. However, the simplest interpretation of the virtual interactions implied by the higher order BDIMs seems to be that these reflect differential tendencies of genes to form paralogous families of different sizes depending on the intensity of selection. Recent studies have shown that evolutionary fixation of gene duplications is linked to the evolutionary rates of genes. Specifically, duplications of slowly evolving genes, i.e., those that are subject to stronger purifying selection, are fixed more often \[[@B54],[@B55]\]. The strong dependence of per gene duplication rates on family size in higher order BDIMs could be an abstraction of this trend. Should that be the case, we are justified to conclude that very weak selection would suffice to explain the stationary distribution of family sizes, but much stronger selective pressure is needed to account for the dynamics of genome evolution. However, the interpretation of BDIM degree as a manifestation of selection is, at this point, no more than a guess. One of the further developments of genome evolution modeling involves introducing selection explicitly and determining whether the resulting more sophisticated models will be equivalent to the higher order BDIMs explored here. Conclusions =========== In this work, we extended our analysis of stochastic Birth, Death and Innovation Models (BDIMs) of gene family evolution and showed that: • the behavior of logistic BDIMs models, in which birth/death rates are limited for the largest families, is essentially the same as that of previously investigated BDIMs that included no such limitation • the mean time required for the growth of large families is limited by the overall number of duplications and does not increase indefinitely with the increase of the model degree but instead passes through a minimum; even under the best-case scenario, which corresponds to a non-linear rational BDIM with d≈ 2.7, the mean time of the largest family formation is orders of magnitude greater than any realistic estimates based on the timescale of life\'s evolution; • using the embedding chains technique, we estimated the expected number of elementary evolutionary events (gene duplications and deletions) preceding the formation of gene families of the observed size; the mean number of events exceeds the family size by orders of magnitude, suggesting a highly dynamic process of genome evolution; • the variance of the time required for the formation of the largest families is large (coefficient of variation \>\> 1), which means that some families might grow much faster than the mean rate; thus, the minimal time required for family formation is more relevant for a realistic representation of genome evolution than the mean time; • Monte Carlo simulations of family growth from an ensemble of simultaneously evolving singletons show that the time elapsed before the formation of the largest family was much shorter than the estimated mean time and approached realistic values (2.2 ± 0.5 Ga for the non-linear rational BDIM with d≈ 2.2). Contributions of individual authors =================================== GPK developed most of the mathematical formalism and wrote the draft of the mathematical part of the manuscript; YIW performed the imitation modeling and wrote the draft of the corresponding part of the manuscript; FSB derived some of the mathematical statements; EVK contributed to the inception of the work and the formulation of the models, gave the biological interpretation of the results, wrote the background and discussion sections and extensively edited the entire manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 This additional file includes proofs of some of the mathematical statements contained in the main text as well as accessory mathematical formulations. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ The authors thank B.Shraiman and other members of the Computational Biology Program at the Kavli Institute for Theoretical Physics, University of California, Santa Barbara, for helpful discussions. The work of F. Berezovskaya was supported by NSF Grant \#634156.	PubMed Central	2024-06-05T03:55:48.617714	2004-9-9	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523855/", "journal": "BMC Evol Biol. 2004 Sep 9; 4:32", "authors": [ { "first": "Georgy P", "last": "Karev" }, { "first": "Yuri I", "last": "Wolf" }, { "first": "Faina S", "last": "Berezovskaya" }, { "first": "Eugene V", "last": "Koonin" } ] }
PMC523856	Background ========== Highly active antiretroviral therapy (HAART) effectively inhibits human immunodeficiency virus (HIV) replication, but it has been suggested that a combination of HAART and strategies for boosting the immune system would give more effective long-term control of HIV infection \[[@B1],[@B2]\]. Interferon β (IFN-β) is an attractive candidate for such therapy: 1) it is a natural, potent antiviral protein that inhibits HIV at various stages of the viral cycle, from uptake to the release of virus particles \[[@B3]-[@B9]\]; 2) Type I IFNs display immunomodulatory properties that could improve the immune control of HIV replication \[[@B10]-[@B12]\]. During HIV infection, the induction of type I IFN production has been shown to be impaired in T cells and macrophages, which are considered to be the major targets of the virus \[[@B13]-[@B16]\]. However, the use of recombinant IFN in therapeutic strategies is limited by its poor bioavailability and the need for high doses to obtain an antiviral effect, resulting in deleterious side effects \[[@B17]\]. It has been suggested that the efficacy of type I IFNs for the treatment of HIV infection could be increased by developing a gene therapy strategy based on the modified production of IFN-β in genetically engineered lymphocytes \[[@B18]\]. For this purpose, a retroviral vector derived from Moloney murine leukemia virus, in which the human IFN-β coding sequence has been placed under the control of a fragment of the murine H2-K^b^gene promoter, has been used to ensure the continuous generation of low levels of IFN-β in transduced cells \[[@B10],[@B19]\]. The transduction of peripheral blood lymphocytes (PBL) with this vector inhibits HIV replication in vitro and increases the survival of CD4^+^cells in culture. Furthermore, IFN-β production in PBL from HIV-infected donors increases Th1-type cytokine production, improves cytotoxic responses against cells expressing HIV proteins, and the proliferative response to recall antigens \[[@B10],[@B12]\]. These in vitro results have been confirmed in the SCID mouse model of HIV infection \[[@B20]\]. However, as the human-SCID mouse has a number of limitations as a model of AIDS, the efficacy and safety of this strategy should also be evaluated in a more appropriate model, such as macaques infected with simian immunodeficiency virus (SIV). SIV resembles HIV-1 and HIV-2 in its genomic organization and biological properties \[[@B21]\] and systematically causes a disease in macaques that is remarkably similar to AIDS in humans \[[@B22]\]. We have previously shown that PBL obtained from seronegative animals and transduced with a vector carrying the macaque IFN-β coding sequence placed under the control of a 0.6-kb fragment of the murine H2-K^b^gene promoter develop greater resistance to SIVmac251 in vitro \[[@B23]\]. In healthy seronegative macaques, infusion with autologous lymphocytes transduced ex vivo with the vector encoding IFN-β results in approximately 1 transduced cell per thousand peripheral blood mononuclear cells (PBMCs). The genetically modified cells were detected for at least 74 days after infusion, with no major side effects, in these experiments. Following infection with SIVmac251, macaques that had received the IFN-β construct infusion displayed lower peak plasma viral loads during primary infection than did control macaques. No adverse reaction was observed, and these macaques maintained high CD4+ T-lymphocyte counts for at least 478 days \[[@B24]\]. However, a gene therapy strategy for HIV infection would only be possible during the chronic phase of infection. At this stage, the immune system, and particularly CD4^+^T cells -- the major target of our gene therapy approach -- may be strongly affected by the virus. We therefore investigated the safety and efficacy of this strategy in macaques chronically infected with a primary, pathogenic isolate of SIVmac251, but still in an asymptomatic state. The efficacy of our strategy has been examined according to two parameters. The eventual survival advantage of IFN-β transduced cells has been monitored by following the presence of such transduced cells in the blood stream as well as in the lymph nodes of infused macaques. This group of animals was compared to a controlgroup having received cells transduced with a retrovirus carrying a modified version of the MaIFN![](1742-4690-1-29-i2.gif) sequence with a deletion blocking mRNA translation. Animals were subjected to three infusions of autologous T lymphocytes transduced ex vivo with both constructs. The eventual clinical benefits of the presence of IFN-β-transduced cells have been monitored for two years, by examining, in both groups of animals, the absolute number of circulating CD4+ lymphocytes, cell associated viral load and plasma vial load. Results ======= Status of animals before treatment ---------------------------------- The in vivo safety and anti-SIV efficacy of IFN-β-engineered lymphocytes in chronically SIV-infected macaques was assessed by following for two years animals that had received three infusions day 0, day 361 and day 613) of autologous T lymphocytes transduced with a construct encoding IFN-β (macaques IFN1, IFN2, IFN3) or, as a control, with a retrovirus carrying a modified version of the IFN-β that could not generate functional protein (macaques C1, C2, C3). These macaques had been infected with 4 AID~50~of a primary, pathogenic isolate of SIVmac251 more than one year before the start of the experiment. On day 0, the mean number of circulating CD4^+^T lymphocytes was 767 ± 215 μl, and all animals had detectable SIV provirus in PBMCs (Table [1](#T1){ref-type="table"}). Plasma SIV viremia was low or undetectable in most animals, the detection threshold being 1,500 copies of SIV RNA copies per milliliter of plasma. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Immunological and virological parameters of macaques at day 0 of the experiment. At the onset of the experiment, the six male cynomolgus macaques (Macaca fascicularis) were chronically infected by 4 AID ~50~of a primary and pathogenic isolate of SIVmac251 for more than one year. They have been characterized for their mean number of circulating CD4+ T-lymphocytes, the time after SIV inoculation, and the cellular and plasma SIV viral loads. IFN group is represented by the three macaques that received their own cells transduced by the biologically active construct of IFN-β gene whereas the control group is represented by the three macaques that received their own cells transduced by the control construct. a: Immunophenotyping of Ficoll-purified PBMCs was performed by immunostaining with specific anti-CD4 and anti-CD8 antibodies, and analyzing by flow cytometry. The mean number of circulating CD4+ T-lymphocytes was determined at day 0 post first infusion with five points preceeding the onset of the experiment. b: Cellular viral load was estimated by a quantitative limit dilution nested PCR method allowing specific double amplification of a gagfragment of SIVmac251. Number of proviral copies was estimated by the last dilution that can display, in an agarose gel, a signal amplification. The number of SIVmac251 gag gene copies per 1 mg of DNA, for instance 131300 cells, was then brought back to a number of gene copies per 10^6^cells. c Plasma SIV viral load was determined by the branched-DNA method. ::: Mean number of circulating CD4^+^lymphocytes ^a^ Time after inoculation of SIVmac251 Mean number of SIV proviral DNA copies in PBMCs ^b^(Copies per 10^6^cells) Plasma SIV load ^c^(10^3^copies per ml) --------- ------ -------------------------------------------------- ------------------------------------- ---------------------------------------------------------------------------- ----------------------------------------- IFN IFN1 811 +/- 154 666 3.5 +/- 3.7 40 IFN2 767 +/- 215 686 116.6 +/- 170.0 \<1.5 IFN3 977 +/- 254 1000 0.9 +/- 2.5 20 Control C1 1356 +/- 172 708 3.5 +/- 3.7 \<1.5 C2 777 +/- 189 313 4.3 +/- 3.6 \<1.5 C3 1055 +/- 478 1830 0.8 +/- 0.0 \<1.5 ::: Transduced PBLs --------------- After being transduced with the MFG-K^b^MaIFN-β and MFG-K^b^ΔMaIFN-β constructs, PBLs were readministered to the animal from which they were originally taken. Each macaque was infused with 10^8^to 4 × 10^8^lymphocytes. Semi-quantitative PCR analysis revealed that the mean transduction efficiencies for the transduction of PBL with the MFG-K^b^MaIFN-β and MFG-K^b^ΔMaIFN-β constructs were 10.33 % ± 7.42 % and 17.13 % ± 10.61 %, respectively. The IFN-β-transduced populations were characterized in culture by a low IFN-β production, ranging from 12 to 24 units per 5 × 10^5^cells per 3 days. We previously published that similar rates of Ma IFN-β-transduction results in a signficant reduction of SIVmac251 replication in vitro \[[@B23]\]. Such IFN-β-transduced cells remain detectable in the blood stream 485 days after reinfusion \[[@B24]\]. After the first inoculation (day 0), transduced cells was detected in peripheral blood, with about 10 transduced cells per 10^6^cells, for 14 days in macaque IFN1, and for 29 days in macaques IFN3 and C2 (Table [2](#T2){ref-type="table"}). A transient peak of 1000 and 700 transduced cells per 10^6^circulating cells was observed in macaques C1 and C3, respectively. After completion of the series of infusions, with the last infusion occurring on day 613, transduced cells persisted at a low level (10 transduced cells per 10^6^cells) for only up to 60 days (Table [2](#T2){ref-type="table"}). No transduced cells were detected at any time in the study for macaque IFN2. No significant difference was observed between the two groups of macaques in terms of transduced cell persistence (Table [2](#T2){ref-type="table"}). The frequency of transduced cells was similar for CD4^+^and CD8^+^lymphocytes analyzed on day 673 (data not shown). No retroviral construct was detected in lymph nodes and splenic mononuclear cells. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### In vivo follow up of transduced cells in blood. Absolute number of transduced cells per 10^6^PBMCs were evaluated by semiquantitative PCR amplification of IFN-β transgene in the two groups of animals. For in vivo follow up of transduced cells in blood from macaques, DNA samples of PBMCs were obtained at different dates following infusion of transduced PBL. This table indicates the minimum and maximum number of days following the first infusion of transduced cells in which the construct used was still detectable in PBMCs. Moreover, maximum transduction rate of PBMCs and detection treshold of the PCR method are indicated in the two groups of animals. IFN group is represented by the three macaques that received their own cells transduced by the biologically active construct of IFN-β gene whereas the control group is represented by the three macaques that received their own cells transduced by the control one. The relative intensity of the signals was compared to serial dilutions of lysate derived from plasmid-transfected cells that contained known numbers of IFN-β transgene copy per cell. a Day 0 is the first infusion day, other infusions occured at days 361 and 613. b Absolute number of transduced cells was below 10 per 10^6^PBMCs. ::: ------------------------ ------ ----------------- ------- ----------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- 1st infusion^a^ Days post-1st infusion 0 1 4 8 12 14 20 22 25 29 33 52 64 81 95 IFN IFN1 ND^b^ 10 10 10 10 10 ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ IFN2 ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ IFN3 ND^b^ 10 10 10 10 10 10 10 10 10 ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ Control C1 ND^b^ 10 10 1000 100 10 10 10 10 10 ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ C2 ND^b^ 10 10 10 10 10 10 10 10 10 ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ C3 ND^b^ 10 700 10 10 10 10 10 10 10 ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ 2nd infusion^a^ 3rd infusion^a^ Days post-1st infusion 361 364 368 375 382 431 489 613 618 625 632 673 688 744 IFN IFN1 ND^b^ 10 10 10 ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ IFN2 ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ ND^b^ IFN3 ND^b^ 10 10 10 ND^b^ ND^b^ ND^b^ ND^b^ 10 10 10 10 ND^b^ ND^b^ Control C1 ND^b^ 10 10 10 ND^b^ ND^b^ ND^b^ ND^b^ 10 10 10 10 ND^b^ ND^b^ C2 ND^b^ 20 20 20 ND^b^ ND^b^ ND^b^ ND^b^ 10 10 10 10 ND^b^ ND^b^ C3 ND^b^ 10 10 10 ND^b^ ND^b^ ND^b^ ND^b^ 10 10 10 10 ND^b^ ND^b^ ------------------------ ------ ----------------- ------- ----------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ::: Clinical status and immunological follow-up ------------------------------------------- Weight and rectal temperature remained fairly constant throughout the study (data not shown). No major variation in classical hematological parameters, including total lymphocyte and platelets counts, and hemoglobin concentration, was observed (data not shown). Immunological follow-up indicated that seven days after the first infusion (day 0), the number of circulating CD4^+^lymphocytes significantly increased in all macaques studied (p \< 0.05), except for C3. A similar significant increase (p \< 0.05) was observed in the days following the second infusion (on day 311) for macaques IFN2, and C1, and following the third infusion (on day 613) for macaques IFN1, and C1 (Fig. [1A](#F1){ref-type="fig"} -- [2A](#F2){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Evolution of immuno-virological parameters in SIVmac251 chronically infected macaques from the IFN group. Immunological and virological parameters were followed in macaques that received their own cells transduced by the retroviral construct allowing expression of the biologically active form of IFN-β. (A) Absolute number of circulating CD4+ lymphocytes was followed by immunophenotyping and flow cytometry. (B) Cell-associated viral load was estimated in PBMCs by a quantitative PCR method based on the specific amplification of the SIV gag gene. (C) Plasma viral load was estimated by a quantitative branched-DNA method based on the specific amplification of the SIV genome. Y axis split X axis at the first reinfusion date (D0) whereas black arrows indicate the second and third reinfusion dates. ::: ![](1742-4690-1-29-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Evolution of immuno-virological parameters in SIVmac251 chronically infected macaques from the control group. Immunological and virological parameters were followed in macaques that received their own cells transduced by the deleted form of the retroviral construct. (A) Absolute number of circulating CD4+ lymphocytes was followed by immunophenotyping and flow cytometry. (B) Cell-associated viral load was estimated in PBMCs by a quantitative PCR method based on the specific amplification of the SIV gag gene. (C) Plasma viral load was estimated by a quantitative branched-DNA method based on the specific amplification of the SIV genome. Y axis split X axis at the first reinfusion date (D0) whereas black arrows indicate the second and third reinfusion dates. ::: ![](1742-4690-1-29-2) ::: For all animals in both groups, absolute numbers of CD4+ T cells gradually decreased during the study (p \< 0.05), and no significant difference in absolute numbers of CD4+ T cells was observed between the two groups of macaques (Fig. [1A](#F1){ref-type="fig"} -- [2A](#F2){ref-type="fig"}). The absolute numbers of circulating CD8^+^T lymphocytes increased (p \< 0.05) transiently during the days following each infusion of transduced cells, in both groups of macaques. However, with the exception of these peaks, absolute numbers of circulating CD8^+^T lymphocytes did not change significantly during the study in any of the animals of either group (data not shown). Virological follow-up of animals -------------------------------- We studied the course of SIV infection by determining the number of copies of SIV proviral DNA per cell, and the number of copies of SIV viral RNA per ml of blood. SIV provirus was detected in the PBMCs of all animals in both groups throughout the study. A transient and significant (p \< 0.05) increase in cellular viral load was observed one to three weeks after each infusion in macaque IFN1 (Fig. [1B](#F1){ref-type="fig"}) and in macaques C1, C2 and C3 (Fig. [2B](#F2){ref-type="fig"}). A similar transient and significant (p \< 0.05) increase in cellular viral load also occurred one to three weeks after the first and second infusions in macaque IFN2 and after the third infusion in macaque IFN3 (Fig. [1A](#F1){ref-type="fig"}). Analysis of the number of SIV RNA copies in the plasma revealed that plasma viremia peaked (p \< 0.05) one week after the first and the third infusions in macaque IFN1 (Fig. [1C](#F1){ref-type="fig"}), after the first infusion in macaque IFN3 (Fig. [1C](#F1){ref-type="fig"}) and after the second infusion in macaque C3, (Fig. [2C](#F2){ref-type="fig"}). The other animals displayed no significant change in plasma viral load during the course of the experiment. Discussion ========== In this study, we assessed the feasibility and efficacy of a gene therapy method based on the introduction into PBL of an IFN-β gene resulting in the constitutive production of low levels of IFN-β, in macaques chronically infected with SIVmac251. The present work was unable to bring new lighting on the efficacy of our gene therapy method since we encountered the problem of disappearence of transduced cells (control or IFN-β transduced cells) few days after each infusion. Throughout the study, significant, transient peaks of cell-associated and / or plasma viral loads were observed in most animals a few weeks after the infusion of transduced cells. These variations may reflect in vivo activation of viral replication, probably due to the infusion of activated cells. This phenomenon was also observed after the infusion into SCID mice of transduced human PBLs, resulting in up-regulation of CCR-5 HIV co-receptor expression in human CD4^+^T cells \[[@B27]\]. Indeed, the SIVmac251 isolate used in our experiment is a CCR5-dependent virus, and its replication may have been activated by upregulation of the CCR-5 coreceptor after infusion. However, gene therapy strategies for the treatment of HIV infection could only be envisaged in combination with HAART. In this context, the activation of host virus replication observed after the infusion of transduced cells would be overcome by HAART treatment. The mean rates of transduction of PBL isolated from macaques chronically infected with SIVmac251 were 10.33 % ± 7.42 % and 17.13 % ± 10.61 % for the MFG-K^b^MaIFN-β and MFG-K^b^ΔMaIFN-β constructs, respectively, which is similar to the transduction efficiency previously reported for PBLs isolated from healthy non infected macaques \[[@B23],[@B24]\]. The transduction efficacy for lymphocytes from healthy donors and HIV-seropositive patients has also been found to be similar \[[@B10]\], indicating that chronic infection does not affect the retroviral transduction of lymphocytes. After the first infusion, small numbers of engineered cells (control and IFN-β-transduced cells) were detected for only 29 days. Thus, the persistence of transduced cells in chronically infected macaques was lower than that previously reported in non infected macaques, in which IFN-β-engineered cells were detected for at least 70 days, and for more than a year after SIVmac251challenge \[[@B24]\]. This former study indicates also that immune response that may be induced by mouse cell components or FCS present in culture medium may not alter persistence of genetically modified immune cells. We carried out three infusions of engineered cells and, after each infusion, the engineered cells disappeared from the bloodstream within a few days. Poor persistence of circulating engineered cells has been reported in HIV-infected macaques and in SCID mice, and has been attributed to the delocalization of circulating transduced cells in the lymph nodes \[[@B28]\], and intestine \[[@B29]\]. In our study, we detected no engineered cells in the lymph nodes or spleen, indicating that the delocalization of transduced cells to these organs could not account for the absence of transduced cells in the blood. The short-term persistence of transduced cells has already been reported in other studies in which autologous engineered T cells were cleared rapidly from the bloodstream \[[@B30]\]. However another group reported the persistence of engineered cells for more than 25 weeks (0.1 to 10% of PBMC) in HIV-infected patients \[[@B29],[@B31],[@B32]\]. They hypothesized that the higher rate of T-cell survival was due to ex vivostimulation through CD3 and CD28. Indeed, it has been demonstrated that the inhibition of HIV replication in CD3- CD28- stimulated CD4^+^cells is due to the production of cytokines associated with Th-1 function \[[@B33]\] and to the downregulation of CCR-5 expression \[[@B34]\]. Thus, in our study, the disappearance of transduced cells may be due to ConA-stimulation, which may induce apoptosis in lymphocytes, as previously described \[[@B35]\]. IFN-β-producing cells and cells transduced with the control vector displayed similar levels of in vivo persistence. We previously reported higher levels of resistance to HIV in vitro following the transduction of human CD4^+^T cells \[[@B19]\], human macrophages \[[@B36]\] and macaque PBL \[[@B23]\] with a construct encoding IFN-β. However, Vieillard et al. \[[@B10]\] reported inefficient protection of transduced lymphocytes against HIV replication in vitro for PBLs isolated from patients in an advanced state of HIV infection. This lack of protection probably resulted from the downregulation of interferon alpha/beta receptor expression in donors with AIDS, leading to hyporesponsiveness to type I IFN \[[@B37]\]. Thus, although we selected animals with CD4^+^cell counts that were still high, the disease may have been so advanced that transducing PBLs with a construct encoding IFN-β had little effect, with the engineered lymphocytes subjected to the high rate of lymphocyte turnover observed during SIV infection \[[@B38],[@B39]\]. Our previous work with the macaque model encouraged us to develop low-level autocrine IFN-β production as an approach to gene therapy for AIDS. The persistence of 1 transduced cell per 10^3^circulating cells before SIV challenge was correlated with low plasma virus load and the maintenance of CD4^+^and CD8^+^cell counts in macaques infused with the construct encoding IFN-β \[[@B24]\]. In this study, performed with animals infected for more than one year, cells transduced with the IFN-β construct rapidly disappeared from the bloodstream after infusion. This suggests that gene therapy by PBL transduction should be performed as soon as possible after primary infection. We are well aware that the number of transduced lymphocytes was too small for a major effect in this study and we believe that further exploration of IFN-β-based anti-HIV therapy will require the construction of high-titer vectors, with the aim of increasing the proportion of vector-transduced HIV target cells. An alternative method for IFN-β gene therapy involves the transduction of CD34^+^hematopoietic stem cells. This method has been proposed for the treatment of HIV infection \[[@B40],[@B41]\]. The transduction of these cells, which are able to generate all the main HIV target cells, will increase the proportion of transduced cells, extend IFN-β production to macrophages and dendritic cells, and should facilitate long-term expression of the therapeutic construct. We have already demonstrated that macrophages transduced with an IFN-β construct display enhanced HIV resistance, and that HIV transmission to CD4^+^T cells is prevented in IFN-β-transduced dendritic cells \[[@B42]\]. We intend to investigate the possibility of transducing hematopoietic stem cells to inhibit viral replication in macaques chronically infected with SIVmac251, in the near future. Methods ======= Animals ------- Six male cynomolgus macaques (Macaca fascicularis), weighing between 3 and 7 kg, and negative for herpes B, filovirus, STLV-1, SRV-1, SRV-2, SIV, and hepatitis-B were used in this study. Before all experimental procedures, animals were anesthetized with chlorhydrate ketamine (Cenravet, France), and all procedures were conducted according to European guidelines for animal care (Official Journal of the European Communities L538, 18 December 1986). Macaques were housed in individual cages in biosafety level 3 facilities, as required by national regulations (Commission de Génie Génétique, Paris, France). Viral stock ----------- More than 300 days before infusion with the IFN construct, macaques were intravenously infected with 4 AID~50~of a primary, pathogenic SIVmac251 isolate. This virus stock was obtained by coculturing splenocytes obtained from an infected rhesus macaque with rhesus macaque PBMCs (Dr. R.C. Desrosiers, Harvard Medical School, MA, USA), and was amplified by a second passage on rhesus PBMCs (prepared and kindly provided by Dr. A.M. Aubertin, Université Louis Pasteur, Strasbourg France). Retroviral vectors ------------------ The MFG-K^b^MaIFN-β retroviral vector used in this study has been described elsewhere \[[@B23]\]. It contains the macaque IFN-β coding sequence placed under the control of a 0.6 kb fragment of the murine H2-K^b^gene promoter, resulting in the continuous production of low levels of a biologically active macaque IFN-β. The MFG-K^b^ΔMaIFN-β retroviral vector used in this study as a control has been described elsewhere \[[@B23]\]. It contains a macaque IFN-β coding sequence with a 530 bp deletion, blocking IFN-β translation, under the control of the same promoter region. Vectors (MFG-K^b^MaIFN-β and MFG-K^b^ΔMaIFN-β were produced with two Ψ-CRIP packaging clones, each of which produced 2 × 10^5^infectious particles per ml, with no detectable replication-competent helper virus \[[@B23]\]. The Ψ-CRIP cells were maintained in Dulbecco\'s modified Eagle\'s medium (DMEM, InVitrogen, Grand Island, New York, USA) supplemented with 10 % heat-inactivated bovine serum (BS) (InVitrogen) and 0.2 μM antibiotics (penicillin / streptomycin / neomycin, PSN, InVitrogen). Isolation of macaque peripheral blood lymphocytes (PBL) ------------------------------------------------------- Three macaques (IFN1, IFN2 and IFN3) received infusions of their own lymphocytes transduced with the biologically active MaIFN-β construct. Another three macaques (C1, C2, C3) were infused with their own lymphocytes transduced with the construct carrying the deleted form of the MaIFN-β, which cannot produce a translatable mRNA. We collected about 100 ml of blood from each macaque into heparin lithium tubes (Greiner, USA). Buffy coats were obtained by centrifugation (170 g / 15 min). Mononuclear cells were collected, and centrifuged (400 g / 30 min) on a Ficoll density gradient (Eurobio, Les Ulis, France). Plasma and erythrocytes, diluted 1 in 2 with 0.9% NaCl (InVitrogen), were washed and used immediately for infusion into the macaques. Transduction of macaque PBLs ---------------------------- Isolated PBMCs (10^6^cells per ml) were activated by incubation for three days in RPMI-1640 medium, 10 % fetal calf serum (FCS), 2 mM L-glutamine (Bœhringer Mannheim, Mannheim, Germany), 0.2 μM antibiotics (penicillin / streptomycin / neomycin), 5 μg / ml concanavalin A (InVitrogen). Activated PBL were resuspended in transduction medium consisting ofn 45 % DMEM, 45 % IMDM (InVitrogen), 5 % FCS, 5 % BS, 4 μg / ml protamine sulfate (Sigma, Saint Louis, USA) and 20 IU / ml recombinant human (rHu) IL-2 (Bœhringer Mannheim). Cells were transduced by coculture for three days with subconfluent Ψ-CRIP packaging cells. At the end of the coculture period, the various cell populations were transferred twice to other culture plates to eliminate any residual adherent packaging cells. Transduced lymphocytes were washed, resuspended in 1× PBS at a concentration of 10^7^cells / ml, and injected intravenously into macaques. Transduction efficacy was estimated with transduced PBLs maintained in culture for 3 days. Evaluation of the transduction rate ----------------------------------- DNA was extracted from macaque PBMCs and the amount used for each sample was normalized based on data for amplification of the β-globin gene, using 5\'-ACCATGGTGCTGTCTCCTGC-3\' as sense primer, and 5\'-CATGGCCACGAGGCTCCA-3\' as an antisense primer. Both retroviral sequences were detected, using 5\'-GTTCAGGCAAAGTCTTAGTC-3\' as the sense primer, binding in the H2-K^b^gene promoter and 5\'-TGAAGATCTCCTAGCCTGT-3 as the antisense primer, binding in the macaque IFN-β coding sequence. These primers amplified a 870-bp fragment from the MFG-K^b^MaIFN-β vector, and a 340-bp fragment from the MFG-K^b^ΔMaIFN-![](1742-4690-1-29-i2.gif) vector The PCR amplification products were identified by dot-blot hybridization with an IFN-β probe, and quantified with a PhosphorImager (Molecular Dynamics, Sevenoaks, England, UK), as previously described \[[@B19]\]. Relative signal intensity was compared with the signal intensity of serial dilutions of lysate derived from plasmid-transfected cells containing known numbers of transgene copies per cell. The detection threshold of the PCR assay used was estimated and found to be one copy of the IFN-β gene per 10^5^cells. Hematological and immunological follow-up of infused macaques ------------------------------------------------------------- All infused animals were followed during the months preceding the study, and for more than 700 days after the first autologous infusion. We carried out hematological analysis, and monitored weight, rectal temperature, and levels of lymphocytes transduced with the IFN-β construct. Blood formula and blood cell counts were determined with an automated hemocytometer (Coulter Corporation, Miami, USA). Axillary lymph nodes and spleens were removed from animals and ground in 1× PBS using a Potter homogenizer. Lymph nodes and splenic mononuclear cells (LNMC, SMC) were then collected and centrifuged (400 g / 30 min) on a Ficoll cushion (Eurobio, Les Ulis, France). DNA extraction and evaluation of in the rate of transduction of LNMC and SMC were performed as previously described. In vivo immunological follow-up of macaques receiving infusions --------------------------------------------------------------- We estimated the proportions of the various subtypes of circulating PBMCs by direct immunofluorescence assay (anti-CD3 clone FN18, Biosource International, CA, USA), anti-CD4 clone Leu 3a PE (Becton Dickinson, San Jose, Mountain View, CA, USA), anti-CD8 clone Leu 2a FITC (Becton Dickinson) antibodies and IgG isotypic controls (Immunotech, Marseille, France), and flow cytometry (Becton Dickinson). We used specific software (CellQuest, Becton Dickinson) as previously described \[[@B25]\] for the analysis. Sorting of CD4^+^and CD8^+^circulating lymphocytes -------------------------------------------------- Mononuclear cells isolated on Ficoll-Hypaque were positively separated using CD4-specific and CD8-specific immunomagnetic microbeads (MiniMACS, Miltenyi, Stadt, Germany) according to manufacturer\'s instructions. Subset purity was evaluated by flow cytometry, using secondary anti-CD4 clone OKT4-PE (Dako, Glostrup, Denmark) and anti-CD8 clone DK25-FITC (Dako) antibodies. The rates of transduction of the sorted CD4^+^and CD8^+^lymphocytes were evaluated, as described above. Plasma and cell-associated viral load ------------------------------------- Levels of SIV RNA in plasma were determined with the SIVmac-branched-DNA assay, using a detection threshold of 1,500 mEq per milliliter of plasma (Chiron Diagnostics, Amsterdam, The Netherlands). DNA was extracted from PBMCs with an extraction kit (Roche Diagnostics GmbH, Mannheim, Germany). Levels of SIV DNA in cells were determined using a two-step PCR method with two external gag-specific primers (1386-5\': GAAACTATGCCAAAAACAAGT and 2129-5\': TAATCTAGCCTTCTGTCCTGG) and two internal gag-specific primers (1731N 5\': CCGTCAGGATCAGATATTGCAGGAA and 2042C 5\': CACTAGCTTGCAATCTGGGTT), as previously described \[[@B26]\]. Statistical analysis -------------------- Statistical significance was determined by paired or unpaired non parametric Wilcoxon and Mann-Whitney tests adapted for small samples. Competing interests =================== The authors never received reimbursements, fees, funding, or salary from an organization that may in any way gain or lose financially from the publication of this paper in the past five years. The authors never any stocks or shares in an organization that may in any way gain or lose financially from the publication of this paper. The authors never have any other financial competing interests. The authors have no non-financial competing interests to declare in relation to this paper. Authors\' contributions ======================= WG was the major contributor to this paper. EL participated in the design of the study and performed the cell cultures and transduction experiments. BB and JL participated in the animals manipulation. FM participated in the preliminary experiments. SP performed all PCR reaction for transduced cells in vivofollow-up. DD and EDM participated in the design and the coordination of the study. RLG performed the statistical analysis and participated in the design and the coordination of the study. Acknowledgements ================ We would like to thank B. Delache, C. Aubenque, P. Brochard, D. Renault, P. Pochard and J.C. Wilk for excellent technical assistance.	PubMed Central	2024-06-05T03:55:48.625436	2004-9-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523856/", "journal": "Retrovirology. 2004 Sep 25; 1:29", "authors": [ { "first": "Wilfried", "last": "Gay" }, { "first": "Evelyne", "last": "Lauret" }, { "first": "Bertrand", "last": "Boson" }, { "first": "Jérome", "last": "Larghero" }, { "first": "Franck", "last": "Matheux" }, { "first": "Sophie", "last": "Peyramaure" }, { "first": "Véronique", "last": "Rousseau" }, { "first": "Dominique", "last": "Dormont" }, { "first": "Edward", "last": "De Maeyer" }, { "first": "Roger", "last": "Le Grand" } ] }
PMC523857	Cloning and characterization of human apoM ========================================== Human apolipoprotein M (apoM) was found and initially isolated from chylomicrons by Xu and Dahlbäck in 1999 \[[@B1]\]. When they performed SDS-PAGE for delipidated human triglyceride-rich lipoprotein (TGRLP) and sequenced protein bands ranging from 6--45 kDa, one of sequences identified as the N-terminal sequence of MFHQIWAALLYFYGI. No homologous protein was identified in public databases, but several human expressed sequence tags (EST) were found similar to these N-terminal amino acid sequence. Based on these sequences, full-length cDNA of the novel protein was obtained with 188 amino acids \[[@B1]\]. Rabbit antibodies were raised against five synthetic peptides based on the protein sequence. The pooled antisera were used to analyze distribution of the protein among various lipoprotein subclasses using Western blotting. Under reducing conditions, a 26-kDa band was particularly predominant in high density lipoprotein (HDL) but was also observed in low density lipoprotein (LDL) and TGRLP. A less pronounced band (approximately 23-kDa) was observed, which corresponded in size to the non-glycosylated protein \[[@B1]\]. As majority of the protein is associated with lipoprotein in plasma, it fulfills the criteria for classification as an apolipoprotein. And this novel protein was named apolipoprotein M (apoM) \[[@B1]\] as the last previously identified apolipoprotein was called apoL \[[@B2]\]. Gel filtration of plasma showed that apoM was associated with HDL-sized particles in wide-type and apoA-I deficient mice and with HDL- and LDL-sized particles in LDL receptor-deficient mice, whereas it was mainly found in VLDL (very low density lipoprotein)-sized particles in high-fat, high-cholesterol-fed apoE deficient mice \[[@B3]\]. These data suggest that apoM mainly associates with HDL in normal mice, but also with the pathologically increased lipoprotein fraction in genetically modified mice. Gene location and amino acid sequence of apoM ============================================= The identified human apoM cDNA (734 base pairs) encoded 188-amino acid residue-long protein. The 5\'-untranslated region was 33 nucleotides and the 3\'-untranslated region 120 nucleotides, not including the poly (A) tail. Southern blot analysis of different species gave positive signals in all mammalian genomes but not in DNA from chicken and yeast \[[@B1]\]. Human apoM gene is located in p21.31 on chromosome 6 (Fig. [1](#F1){ref-type="fig"}) (chromosome 17, in mouse). The genomic sequence of this region was determined and the human apoM gene identified (GenBank accession number AF118393). In human genome, the apoM gene is surrounded by BAT4 and NG34 on one side and BAT3 on the other. Both mouse apoM gene and its human counterpart are predicted to contain 6 exons enclosed in a 1.6-kb genomic region, which is consistent with the results of Southern blotting. The calculated molecular mass of the protein was 21,256. There is one potential site for N-linked glycosylation at Asn-135 (Asn-Glu-Thr), whereas Asn-148 (Asn-Arg-Ser-Pro) is less likely to be glycosylated because Pro-151 follows Ser-150. The amino acid sequences of human and mouse apoM are 79% identical (82%, human and rat apoM) (Fig. [2](#F2){ref-type="fig"}), and just like human apoM the mouse sequence predicts the presence of a signal anchor, as there is no predicted signal peptidase cleavage site. The amino acid sequence of apoM contained six cysteines, which may involve in the formation of three disulfide bridges. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### ApoM gene location in chromosome 6. ApoM gene is located in chromosome 6 p21.31 <http://bioinfo.weizmann.ac.il/cards-bin/carddisp?APOM>. ::: ![](1476-511X-3-21-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Comparison of apoM amino acid sequence of rat, human and mouse. Dots indicate residues that are identical to the top line (rat). One potential site for N-linked glycosylation site (Asn-Glu-Thr) is indicated by a large dot above the sequence of human apoM (○). Two sequences underlined indicate the typical lipocalin motifs. ::: ![](1476-511X-3-21-2) ::: Protein structure of apoM ========================= Based on sensitive sequence searches, it is proposed that apoM is related to the lipocalin protein superfamily (Fig. [2](#F2){ref-type="fig"}) \[[@B1]\]. Subsequently, Duan et al. \[[@B4]\], used computer protein modeling of two lipocalins, mouse major urinary protein (MUP) and human retinol binding protein (RBP) as initial templates to build apoM protein structure, which demonstrated that apoM has the same structure of lipicalin protein superfamily. ApoM retains an uncleaved N-terminal signal peptide that most likely anchors the molecule into single layer lipids on HDL \[[@B4]\]. The major phospholipids in HDL is phosphatidylcholine, which has a positively charged choline group exposed to the solvent. Two electronegative regions are striking in the apoM model and are located around the N-terminus and the opening of the binding pocket. In this three-dimensional model, characterized by an eight-stranded anti-parallel β-barrel, a segment including Asn-135 could adopt a closed or open conformation. ApoM presents three disulfide bridges, which would make it a member of the lipcalin subgroup of proteins with three s-s bonds \[[@B4]\]. Tissue distribution and cellular expression of apoM =================================================== Northern blot analyses of multiple tissues (including spleen, thymus, prostate, testis, ovary, small intestine, colon, leukocytes, heart, brain, placenta, lung, liver, skeletal muscle, kidney, pancreas, stomach, thyroid, spinal cord, lymph node, trachea, adrenal gland and bone marrow) showed that apoM was mainly expressed in kidney and liver \[[@B1]\]. Furthermore, human tissue expression array study indicated that apoM is only expressed in liver and in kidney and small amounts were found in fetal liver and in fetal kidney \[[@B5]\]. To elucidate whether and when apoM is expressed, Zhang et al. investigated apoM expression patterns during mouse and human embryogenesis \[[@B6]\]. ApoM transcripts were detectable in mouse embryos day 7.5 to day 18.5. It was expressed at low levels at day 7.5, increased significantly at day 9.7 and decreased at day 10.5, and then increased continually up to day 18.5. ApoM-positive cells appeared mainly in liver of day 12 embryos as detected by in situhybridization. In day-15 embryos, apoM was expressed in both liver and kidney. During human embryogenesis, apoM was strongly expressed in livers of 3--5 month-old human embryos and continued to be strongly expressed throughout embryogenesis. In the kidney, apoM expression was highest in 5--9 month-old embryos. There was some expression of apoM in small intestine, particularly in later stages of embryogenesis. In skeletal muscle, minute apoM expression was found in 3--5 months-old embryos, and some apoM expression was found in stomach in earlier stages of embryogenesis \[[@B6]\]. These finding suggest that apoM has high organ specificity and strongly indicate that the physiological function of apoM must be related with liver and kidney. Both immunohistochemical staining and in situapoM mRNA hybridization demonstrated that apoM is exclusively expressed in the hepatocytes and in the tubule epithelial cells in human kidney \[[@B5]\]. Thus, apoM may have specific function in vivo, which may be related to the hepatic lipid and/or lipoprotein metabolism. Regulation of apoM expression ============================= In vitro, several biological factors have been tested to examine their influences over the transcription and secretion of apoM in hepatic cell line (HepG2 cells). Like apoB, apoM is highly hydrophobic and must co-circulate with lipoprotein particles in the blood stream. It has been demonstrated that apoB could be down-regulated by transforming growth factor-beta (TGF-β) \[[@B7],[@B8]\]. Xu et al reported that TGF-β could also down-regulate apoM expression and secretion in HepG2 cells \[[@B9]\]. It suggests that apoM, similar to apoB, may involve in the hepatic lipoprotein metabolism in vivo. In another study, Xu et al demonstrated that platelet-activating factor (PAF) could up-regulate apoM expression in HepG2 cells, whereas, lexipafant, a PAF-receptor antagonist significantly suppressed the mRNA levels and the secretion of apoM in HepG2 cells in a dose-dependent manner. Neither tumor necrosis factor-α (TNF-α) nor interleukin-1α (IL-1α) influences apoM expression or secretion in HepG2 cell cultures \[[@B10]\]. It indicates that apoM may relate to the host defense response because apoM gene is located in histocompatibility complex III (HMC-III) region on chromosome 6. Many genes in this region are related to the immune response, and the apoM gene is very close to the TNF-α gene and lymphotoxin genes. Thus, apoM may also be related to the immune response system, or regulated by cytokines or other inflammatory factors. Administration of adrenocorticotropic hormone (ACTH) has beneficial effects on plasma lipoproteins \[[@B11]-[@B15]\]. A consistent decrease of plasma total cholesterol and LDL cholesterol by 20--40% is seen during ACTH treatment \[[@B14]-[@B17]\]. It has been demonstrated that pronounced hypolipidimic effects of ACTH might be related to the inhibition of apoB synthesis in hepatic cells \[[@B18]\]. However ACTH didn\'t influence apoM expression and secretion in vivoand in vitro\[[@B18],[@B19]\], indicates that apoM may have somewhat difference from apoB on lipid and/or lipoprotein metabolism in vivo. Richter et al. reported that apoM gene expression could be regulated by HNF-1α. Mutant HNF-1α-/- mice completely lack expression of apoM in liver and kidney. Serum apoM levels in HNF-1α+/- mice are reduced by 50% compared with wild-type animals. By analyzing the apoM promoter and identifying a conserved HNF-1 binding site, they showed that HNF-1α is a potent activator of apoM promoter, that a specific mutation in the HNF-1 binding site abolished transcriptional activation of apoM gene. HNF-1α protein can bind to the HNF-1 binding site of apoM promoter in vitro\[[@B20]\]. Liang and Tall reported that leptin up-regulated mRNA level of apoM in ob/obmice \[[@B21]\], suggesting that leptin could stimulate hepatic cells to produce apoM. Faber et al. reported that plasma concentration of apoM was similar in wild-type, LDL receptor-deficient and apoE deficient mice but was reduced by 33% in apoA-I-deficient mice compared with the wide-type mice, which suggest a connection between apoM and apoA-I metabolism \[[@B3]\]. Xu et al., found that in both liver and kidney, expression of apoM was significantly lower in leptin deficient ob/obmice and in leptin-receptor deficient db/dbmice than in control mice. Furthermore, leptin administration significantly increased plasma apoM levels and apoM mRNA levels in liver and in kidney in ob/obmice \[[@B22]\]. It is concluded that both leptin and leptin-receptor are essential for the apoM expression, indicating that leptin is a physiological regulating factor on apoM synthesis in vivo. Physiopathology and potential clinical importance of apoM ========================================================= Xu et al. investigated the relationship between plasma apoM levels and leptin levels, body mass index (BMI), fasting glucose, fasting insulin as well as lipoprotein concentrations in females displaying a wide range in BMI (18.9--57.1 kg/m^2^, n = 51). In univariate analysis, apoM correlated significantly to leptin (r = 0.54, P \< 0.001), BMI (r = 0.70, P \< 0.001), fasting insulin (r = 0.33, P = 0.025), total cholesterol (r = -0.41, P = 0.016), and LDL-cholesterol (r = -0.39, P = 0.018). The correlations between apoM and cholesterol and between apoM and leptin remained significant after adjustment for the influence of BMI. Forward stepwise multiple regressions when leptin, BMI, insulin and cholesterol were entered in a model as independent variables and apoM as the dependent variable showed that cholesterol and leptin were independent predictors of circulating apoM. These two parameters yielded an r^2^of 0.28, thereby explaining approximately 30% of the variance in apoM. Hence, apoM is positively correlated to leptin and negatively correlated to cholesterol levels in humans \[[@B23]\]. Richter et al. measured apoM levels in the serum of nine HNF-1α /maturity-onset diabetes of the young (MODY3) patients, nine normal matched control subjects (HNF-1α +/+), and nine HNF-4α /MODY1 subjects. Serum levels of apoM were significantly decreased in HNF-1α /MODY3 subjects when compared with control subjects as well as with HNF-4α /MODY1 subjects, indicating that HNF-1α haploinsufficiency rather than hyperglycemia is the primary cause of decreased serum apoM protein concentrations. Thus, serum levels of apoM may be a useful serum marker for the identification of MODY3 patients \[[@B20]\]. Alzheimer\'s disease (AD) is a complex, multifactor disorder, probably resulting from an interaction between environmental and genetic factors \[[@B24]-[@B26]\]. Increasing evidence points to a link between cholesterol turnover and AD \[[@B27],[@B28]\], suggesting that genes implicated in brain cholesterol homeostasis may be potential candidate genes for AD. It is well known that apoE genotype and apoE receptor are related to AD \[[@B28],[@B29]\]. With this background, Kabbara et al examined association of apoM with the risk of developing AD. It is excluded apoM as a genetic determinant of AD in a large French case control population \[[@B30]\]. Conclusion ========== In conclusion, apoM is a novel HDL apolipoprotein. Like apoB apoM could be regulated by several cytokines in vivoand in vitro. HNF-1α is one of the most important activator of apoM gene promoter. Plasma apoM concentration is positively correlated to leptin levels and negatively related to plasma cholesterol levels. Both leptin and leptin receptor are essential for apoM expression in vivo. Plasma apoM levels may be used as the marker for identification of MODY3. The detailed relationship between apoM and MODY3 as well as obese needs further investigation.	PubMed Central	2024-06-05T03:55:48.628710	2004-10-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523857/", "journal": "Lipids Health Dis. 2004 Oct 4; 3:21", "authors": [ { "first": "Guanghua", "last": "Luo" }, { "first": "Xiaoying", "last": "Zhang" }, { "first": "Peter", "last": "Nilsson-Ehle" }, { "first": "Ning", "last": "Xu" } ] }
PMC523858	Introduction ============ Tuberculosis continues to be a devastatingly important health problem in the world. In addition to pulmonary involvement, extrapulmonary involvement becomes an issue in most developing countries. Extrapulmonary tuberculosis, because of several factors, has greatly contributed to the total tuberculosis mortalities during the 20^th^century \[[@B1]\]. Risk of tuberculosis has increased due to decreased immunity in uremic patients. Tuberculosis comes out extrapulmonary with a rate of 40 percent in these patients, and periton is involved in 6 percent of all cases \[[@B2]\]. The risk increases in hemodialisis patients within 12 months after the beginning of treatment \[[@B3],[@B4]\]. Tuberculous (TB) peritonitis is an event rarely seen in continuous ambulatory peritoneal dialysis CAPD patients \[[@B5]\]. Our case is presented as a rare TB peritonitis event receiving CAPD treatment. Case Report =========== The 36 year- old male patient, after receiving CAPD treatment for 4 months, consulted our clinic because of stomachache, diarrhea, nausea, vomiting and continous fever. The patient had been diagnosed with chronic renal deficiency and had been followed up with diagnosis of predialysis CRF for 4 years. The patient was referred to us because of his symptoms such as of nausea, vomiting, weakness, and a general condition of fatigue. Immediate care involved an urgent hemodialysis followed by CAPD and planning for renal replacement therapy. Through a physical examination, the patient\'s blood pressure was 110/70 mmHg. The general condition was bad and pulmonary sounds in the respiratory system were diminished slightly in lower zones. On CAPD catheter\'s entering segment, infections were not seen. In a laboratory investigation Hg was at 9.1gr/dl, Htc was at %27.6, WBC was at 5200/mm^3^, the platelet count was at 203000/mm^3^, and the erythrocyte sedimentation rate was at 22 mm/h. In biochemical findings, furthermore, serum creatinine was at 6.05 mgr/dl \[ref. 0,5--1,4\], urea nitrogen was at 38 mgr/dl \[ref.5--20\], protein was at 3.8gr/dl \[ref:6--8.5\], albumin was at 1.2 gr/dl \[ref:3,5--5\], and lactic dehidrogenase was at 565 iu/L. Serum sodium, potassium, glucose, bilirubin, alkaline phosphatase, aspartate and alanine aminotransferase, gamma-glutamyl transpeptidase, amylase, triglyseride, and cholesterol were normal. A coagulation factor protrombin time was found to be 18.7 sn. C-reactive protein was 9.54 mgr/dl. Bilateral costofrenic angles were blunted in posteroanterior pulmonary graphy. No parasites and cystes were found in fecal examination due to diarrhea. No pathogenic agent was detected in stool cultures. In peritoneal cell counting, 1600/mm^3^cell were detected and it was seen that 70 percent of these cells were polymorphonuclear leukocytosis (PMNL). The patient was given ceftazidime (IV), cephazol, and amikacin (intraperitoneal), but no benefit was noticed after 12 days of antibiotherapy and there was no growth in peritoneal fluid cultures. There were PMNL present but no microorganism could be detected. Acid-fast basilli (AFB) was found to be positive in the gram staining of peritoneal fluid in the remaining follow up periods, and the patient had begun antituberculosis therapy in fours(with isoniazid, rifampin, ethambutol and pyrazinamide). Tuberculin skin test was anergical. On the 15^th^day of anti tbc therapy, peritoneal fluid cell count decreased to 300/mm^3^. Peritoneal fluid bacterial culture, blood cultures, throat culture and urine culture were negative but peritoneal fluid tbc culture was found to be positive, in Lowenstein- Jensen medium in 24 days. The patient was followed up with the treatment for recovery with an anti-tbc treatment. The peritoneal fluid of the patient was sent to be examined with Gram staining and Ziehl Neelsen staining. The peritoneal fluid was centrifuged at 3,000 × g for 15 minutes and the sediment was stained by Gram and Ziehl-Neelsen staining. The Gram staining showed PMNL presence but no microorganisms. The Ziehl-Neelsen staining(AFB) was positive. The peritoneal fluid was transferred to 10 ml sterile glass tube and centrifuged at 3,000 × g for 15 minutes. The concentrated sediment was inoculated onto Lowenstein Jensen (LJ) medium without prior decontamination. LJ medium was incubated at 37°C. Two specimens were later sent to be examined with Ziehl Neelsen staining on two different days. Both of them were detected to be positive for Ziehl Neelsen staining. LJ medium was examined for growth twice weekly for the first two weeks and once a week thereafter until the eighth week. After 24 days, the colonies were able to be seen on LJ medium. Positive growth was confirmed by Ziehl Neelsen staining. Discussion ========== CRF increases the risk of tuberculosis. In patients receiving hemodialisis, the risk of tbc increases within twelve months after the occurrence of extrapulmonary tbc. The risk in these patients is ten times more for extrapulmonary tbc than in any other population. Peritoneal tuberculosis is rarely seen but remains a very important complication in CAPD patients\[[@B5],[@B6]\]. Mortality is high in these patients \[[@B7]\]. There are literatures showing mortality rates as high as 15 percent \[[@B8]\]. Quantrill at al., in a TB peritonitis study with 8 cases, found bacterial peritonitis as a source of the patient\'s complaints \[[@B5]\]. It was reported that this patient\' acute course was atypical with a predominance of neutrophils and low levels of protein in the peritoneal fluid \[[@B9]\]. In English literature the most common complaints of tbc peritonit are as follows: fever (78 percent), stomachache (92 percent), misty dialisat (90 percent) and PMNL are dominant in peritoneal fluids in 76 percent of the cases and in 73 percent of the cases AFB and culture are positive \[[@B8]\]. Abraham et al. have reported tbc peritonit in 4 of 155 CAPD patients and tuberculin test were found anergical in all patients \[[@B10]\]. In our case the tuberculin test result was found anergical as well. In a retrospective study made by Lui et al. pulmonary or extrapulmonary tbc was detected in 38 of 790 CAPD patients and they obtained benefits on the 7^th^--57^th^days of antituberculosis treatment (on average 30 day) \[[@B11]\]. We had experienced a recession in the peritonitis of the patient after 15th day of antituberculosis treatment. It was reported in the literature; that in tbc peritonitis treatment, removing peritoneal catheter has no apparent benefit and does not increase efficacy of the treatment \[[@B2],[@B6],[@B9]\]. Considering this case, we think that in patients with CAPD catheter and peritonitis; when peritoneal fluid leukocytes are high and PMNL are dominant, AFB and tuberculosis culture must be routinely investigated along with bacterial culture.	PubMed Central	2024-06-05T03:55:48.630298	2004-10-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC523858/", "journal": "Ann Clin Microbiol Antimicrob. 2004 Oct 4; 3:19", "authors": [ { "first": "Garip", "last": "Sahin" }, { "first": "Nuri", "last": "Kiraz" }, { "first": "Ilknur", "last": "Sahin" }, { "first": "Mehmet", "last": "Soydan" }, { "first": "Yurdanur", "last": "Akgün" } ] }

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.